Georgios Tzimiropoulos

Junting Pan, Adrian Bulat, Fuwen Tan et al.

Self-attention based models such as vision transformers (ViTs) have emerged as a very competitive architecture alternative to convolutional neural networks (CNNs) in computer vision. Despite increasingly stronger variants with ever-higher recognition accuracies, due to the quadratic complexity of self-attention, existing ViTs are typically demanding in computation and model size. Although several successful design choices (e.g., the convolutions and hierarchical multi-stage structure) of prior CNNs have been reintroduced into recent ViTs, they are still not sufficient to meet the limited resource requirements of mobile devices. This motivates a very recent attempt to develop light ViTs based on the state-of-the-art MobileNet-v2, but still leaves a performance gap behind. In this work, pushing further along this under-studied direction we introduce EdgeViTs, a new family of light-weight ViTs that, for the first time, enable attention-based vision models to compete with the best light-weight CNNs in the tradeoff between accuracy and on-device efficiency. This is realized by introducing a highly cost-effective local-global-local (LGL) information exchange bottleneck based on optimal integration of self-attention and convolutions. For device-dedicated evaluation, rather than relying on inaccurate proxies like the number of FLOPs or parameters, we adopt a practical approach of focusing directly on on-device latency and, for the first time, energy efficiency. Specifically, we show that our models are Pareto-optimal when both accuracy-latency and accuracy-energy trade-offs are considered, achieving strict dominance over other ViTs in almost all cases and competing with the most efficient CNNs. Code is available at https://github.com/saic-fi/edgevit.

Jing Yang, Xiatian Zhu, Adrian Bulat et al. · cambridge

Existing knowledge distillation methods mostly focus on distillation of teacher's prediction and intermediate activation. However, the structured representation, which arguably is one of the most critical ingredients of deep models, is largely overlooked. In this work, we propose a novel {\em \modelname{}} ({\bf\em \shortname{})} method dedicated for distilling representational knowledge semantically from a pretrained teacher to a target student. The key idea is that we leverage the teacher's classifier as a semantic critic for evaluating the representations of both teacher and student and distilling the semantic knowledge with high-order structured information over all feature dimensions. This is accomplished by introducing a notion of cross-network logit computed through passing student's representation into teacher's classifier. Further, considering the set of seen classes as a basis for the semantic space in a combinatorial perspective, we scale \shortname{} to unseen classes for enabling effective exploitation of largely available, arbitrary unlabeled training data. At the problem level, this establishes an interesting connection between knowledge distillation with open-set semi-supervised learning (SSL). Extensive experiments show that our \shortname{} outperforms significantly previous state-of-the-art knowledge distillation methods on both coarse object classification and fine face recognition tasks, as well as less studied yet practically crucial binary network distillation. Under more realistic open-set SSL settings we introduce, we reveal that knowledge distillation is generally more effective than existing Out-Of-Distribution (OOD) sample detection, and our proposed \shortname{} is superior over both previous distillation and SSL competitors. The source code is available at \url{https://github.com/jingyang2017/SRD\_ossl}.

Mohammad Mahdi Derakhshani, Enrique Sanchez, Adrian Bulat et al.

Foundational image-language models have generated considerable interest due to their efficient adaptation to downstream tasks by prompt learning. Prompt learning treats part of the language model input as trainable while freezing the rest, and optimizes an Empirical Risk Minimization objective. However, Empirical Risk Minimization is known to suffer from distributional shifts which hurt generalizability to prompts unseen during training. By leveraging the regularization ability of Bayesian methods, we frame prompt learning from the Bayesian perspective and formulate it as a variational inference problem. Our approach regularizes the prompt space, reduces overfitting to the seen prompts and improves the prompt generalization on unseen prompts. Our framework is implemented by modeling the input prompt space in a probabilistic manner, as an a priori distribution which makes our proposal compatible with prompt learning approaches that are unconditional or conditional on the image. We demonstrate empirically on 15 benchmarks that Bayesian prompt learning provides an appropriate coverage of the prompt space, prevents learning spurious features, and exploits transferable invariant features. This results in better generalization of unseen prompts, even across different datasets and domains. Code available at: https://github.com/saic-fi/Bayesian-Prompt-Learning

Stella Bounareli, Christos Tzelepis, Vasileios Argyriou et al.

In this paper, we present our method for neural face reenactment, called HyperReenact, that aims to generate realistic talking head images of a source identity, driven by a target facial pose. Existing state-of-the-art face reenactment methods train controllable generative models that learn to synthesize realistic facial images, yet producing reenacted faces that are prone to significant visual artifacts, especially under the challenging condition of extreme head pose changes, or requiring expensive few-shot fine-tuning to better preserve the source identity characteristics. We propose to address these limitations by leveraging the photorealistic generation ability and the disentangled properties of a pretrained StyleGAN2 generator, by first inverting the real images into its latent space and then using a hypernetwork to perform: (i) refinement of the source identity characteristics and (ii) facial pose re-targeting, eliminating this way the dependence on external editing methods that typically produce artifacts. Our method operates under the one-shot setting (i.e., using a single source frame) and allows for cross-subject reenactment, without requiring any subject-specific fine-tuning. We compare our method both quantitatively and qualitatively against several state-of-the-art techniques on the standard benchmarks of VoxCeleb1 and VoxCeleb2, demonstrating the superiority of our approach in producing artifact-free images, exhibiting remarkable robustness even under extreme head pose changes. We make the code and the pretrained models publicly available at: https://github.com/StelaBou/HyperReenact .

Stella Bounareli, Christos Tzelepis, Vasileios Argyriou et al.

In this paper we address the problem of neural face reenactment, where, given a pair of a source and a target facial image, we need to transfer the target's pose (defined as the head pose and its facial expressions) to the source image, by preserving at the same time the source's identity characteristics (e.g., facial shape, hair style, etc), even in the challenging case where the source and the target faces belong to different identities. In doing so, we address some of the limitations of the state-of-the-art works, namely, a) that they depend on paired training data (i.e., source and target faces have the same identity), b) that they rely on labeled data during inference, and c) that they do not preserve identity in large head pose changes. More specifically, we propose a framework that, using unpaired randomly generated facial images, learns to disentangle the identity characteristics of the face from its pose by incorporating the recently introduced style space $\mathcal{S}$ of StyleGAN2, a latent representation space that exhibits remarkable disentanglement properties. By capitalizing on this, we learn to successfully mix a pair of source and target style codes using supervision from a 3D model. The resulting latent code, that is subsequently used for reenactment, consists of latent units corresponding to the facial pose of the target only and of units corresponding to the identity of the source only, leading to notable improvement in the reenactment performance compared to recent state-of-the-art methods. In comparison to state of the art, we quantitatively and qualitatively show that the proposed method produces higher quality results even on extreme pose variations. Finally, we report results on real images by first embedding them on the latent space of the pretrained generator. We make the code and pretrained models publicly available at: https://github.com/StelaBou/StyleMask

Anurag Das, Adrian Bulat, Alberto Baldrati et al.

Large Vision Language Models (LVLMs) have demonstrated remarkable capabilities, yet their proficiency in understanding and reasoning over multiple images remains largely unexplored. While existing benchmarks have initiated the evaluation of multi-image models, a comprehensive analysis of their core weaknesses and their causes is still lacking. In this work, we introduce MIMIC (Multi-Image Model Insights and Challenges), a new benchmark designed to rigorously evaluate the multi-image capabilities of LVLMs. Using MIMIC, we conduct a series of diagnostic experiments that reveal pervasive issues: LVLMs often fail to aggregate information across images and struggle to track or attend to multiple concepts simultaneously. To address these failures, we propose two novel complementary remedies. On the data side, we present a procedural data-generation strategy that composes single-image annotations into rich, targeted multi-image training examples. On the optimization side, we analyze layer-wise attention patterns and derive an attention-masking scheme tailored for multi-image inputs. Experiments substantially improved cross-image aggregation, while also enhancing performance on existing multi-image benchmarks, outperforming prior state of the art across tasks. Data and code will be made available at https://github.com/anurag-198/MIMIC.

Christos Tzelepis, James Oldfield, Georgios Tzimiropoulos et al.

This work addresses the problem of discovering non-linear interpretable paths in the latent space of pre-trained GANs in a model-agnostic manner. In the proposed method, the discovery is driven by a set of pairs of natural language sentences with contrasting semantics, named semantic dipoles, that serve as the limits of the interpretation that we require by the trainable latent paths to encode. By using the pre-trained CLIP encoder, the sentences are projected into the vision-language space, where they serve as dipoles, and where RBF-based warping functions define a set of non-linear directional paths, one for each semantic dipole, allowing in this way traversals from one semantic pole to the other. By defining an objective that discovers paths in the latent space of GANs that generate changes along the desired paths in the vision-language embedding space, we provide an intuitive way of controlling the underlying generative factors and address some of the limitations of the state-of-the-art works, namely, that a) they are typically tailored to specific GAN architectures (i.e., StyleGAN), b) they disregard the relative position of the manipulated and the original image in the image embedding and the relative position of the image and the text embeddings, and c) they lead to abrupt image manipulations and quickly arrive at regions of low density and, thus, low image quality, providing limited control of the generative factors. We provide extensive qualitative and quantitative results that demonstrate our claims with two pre-trained GANs, and make the code and the pre-trained models publicly available at: https://github.com/chi0tzp/ContraCLIP

Alexandros Xenos, Themos Stafylakis, Ioannis Patras et al.

This paper is on the problem of Knowledge-Based Visual Question Answering (KB-VQA). Recent works have emphasized the significance of incorporating both explicit (through external databases) and implicit (through LLMs) knowledge to answer questions requiring external knowledge effectively. A common limitation of such approaches is that they consist of relatively complicated pipelines and often heavily rely on accessing GPT-3 API. Our main contribution in this paper is to propose a much simpler and readily reproducible pipeline which, in a nutshell, is based on efficient in-context learning by prompting LLaMA (1 and 2) using question-informative captions as contextual information. Contrary to recent approaches, our method is training-free, does not require access to external databases or APIs, and yet achieves state-of-the-art accuracy on the OK-VQA and A-OK-VQA datasets. Finally, we perform several ablation studies to understand important aspects of our method. Our code is publicly available at https://github.com/alexandrosXe/ASimple-Baseline-For-Knowledge-Based-VQA

Debin Meng, Christos Tzelepis, Ioannis Patras et al.

Generating human portraits is a hot topic in the image generation area, e.g. mask-to-face generation and text-to-face generation. However, these unimodal generation methods lack controllability in image generation. Controllability can be enhanced by exploring the advantages and complementarities of various modalities. For instance, we can utilize the advantages of text in controlling diverse attributes and masks in controlling spatial locations. Current state-of-the-art methods in multimodal generation face limitations due to their reliance on extensive hyperparameters, manual operations during the inference stage, substantial computational demands during training and inference, or inability to edit real images. In this paper, we propose a practical framework - MM2Latent - for multimodal image generation and editing. We use StyleGAN2 as our image generator, FaRL for text encoding, and train an autoencoders for spatial modalities like mask, sketch and 3DMM. We propose a strategy that involves training a mapping network to map the multimodal input into the w latent space of StyleGAN. The proposed framework 1) eliminates hyperparameters and manual operations in the inference stage, 2) ensures fast inference speeds, and 3) enables the editing of real images. Extensive experiments demonstrate that our method exhibits superior performance in multimodal image generation, surpassing recent GAN- and diffusion-based methods. Also, it proves effective in multimodal image editing and is faster than GAN- and diffusion-based methods. We make the code publicly available at: https://github.com/Open-Debin/MM2Latent

Ioannis Maniadis Metaxas, Georgios Tzimiropoulos, Ioannis Patras

Clustering has been a major research topic in the field of machine learning, one to which Deep Learning has recently been applied with significant success. However, an aspect of clustering that is not addressed by existing deep clustering methods, is that of efficiently producing multiple, diverse partitionings for a given dataset. This is particularly important, as a diverse set of base clusterings are necessary for consensus clustering, which has been found to produce better and more robust results than relying on a single clustering. To address this gap, we propose DivClust, a diversity controlling loss that can be incorporated into existing deep clustering frameworks to produce multiple clusterings with the desired degree of diversity. We conduct experiments with multiple datasets and deep clustering frameworks and show that: a) our method effectively controls diversity across frameworks and datasets with very small additional computational cost, b) the sets of clusterings learned by DivClust include solutions that significantly outperform single-clustering baselines, and c) using an off-the-shelf consensus clustering algorithm, DivClust produces consensus clustering solutions that consistently outperform single-clustering baselines, effectively improving the performance of the base deep clustering framework.

Adrian Bulat, Georgios Tzimiropoulos

Soft prompt learning has recently emerged as one of the methods of choice for adapting V&L models to a downstream task using a few training examples. However, current methods significantly overfit the training data, suffering from large accuracy degradation when tested on unseen classes from the same domain. To this end, in this paper, we make the following 4 contributions: (1) To alleviate base class overfitting, we propose a novel Language-Aware Soft Prompting (LASP) learning method by means of a text-to-text cross-entropy loss that maximizes the probability of the learned prompts to be correctly classified with respect to pre-defined hand-crafted textual prompts. (2) To increase the representation capacity of the prompts, we propose grouped LASP where each group of prompts is optimized with respect to a separate subset of textual prompts. (3) We identify a visual-language misalignment introduced by prompt learning and LASP, and more importantly, propose a re-calibration mechanism to address it. (4) We show that LASP is inherently amenable to including, during training, virtual classes, i.e. class names for which no visual samples are available, further increasing the robustness of the learned prompts. Through evaluations on 11 datasets, we show that our approach (a) significantly outperforms all prior works on soft prompting, and (b) matches and surpasses, for the first time, the accuracy on novel classes obtained by hand-crafted prompts and CLIP for 8 out of 11 test datasets. Code will be made available at https://www.adrianbulat.com/lasp

Adrian Bulat, Ricardo Guerrero, Brais Martinez et al.

This paper is on Few-Shot Object Detection (FSOD), where given a few templates (examples) depicting a novel class (not seen during training), the goal is to detect all of its occurrences within a set of images. From a practical perspective, an FSOD system must fulfil the following desiderata: (a) it must be used as is, without requiring any fine-tuning at test time, (b) it must be able to process an arbitrary number of novel objects concurrently while supporting an arbitrary number of examples from each class and (c) it must achieve accuracy comparable to a closed system. Towards satisfying (a)-(c), in this work, we make the following contributions: We introduce, for the first time, a simple, yet powerful, few-shot detection transformer (FS-DETR) based on visual prompting that can address both desiderata (a) and (b). Our system builds upon the DETR framework, extending it based on two key ideas: (1) feed the provided visual templates of the novel classes as visual prompts during test time, and (2) ``stamp'' these prompts with pseudo-class embeddings (akin to soft prompting), which are then predicted at the output of the decoder. Importantly, we show that our system is not only more flexible than existing methods, but also, it makes a step towards satisfying desideratum (c). Specifically, it is significantly more accurate than all methods that do not require fine-tuning and even matches and outperforms the current state-of-the-art fine-tuning based methods on the most well-established benchmarks (PASCAL VOC & MSCOCO).

I-Hsiang Chen, Isma Hadji, Enrique Sanchez et al.

Image restoration aims to recover high quality images from inputs degraded by various factors, such as adverse weather, blur, or low light. While recent studies have shown remarkable progress across individual or unified restoration tasks, they still suffer from limited generalization and inefficiency when handling unknown or composite degradations. To address these limitations, we propose RAR, a Restore, Assess and Repeat process, that integrates Image Quality Assessment (IQA) and Image Restoration (IR) into a unified framework to iteratively and efficiently achieve high quality image restoration. Specifically, we introduce a restoration process that operates entirely in the latent domain to jointly perform degradation identification, image restoration, and quality verification. The resulting model is fully trainable end to end and allows for an all-in-one assess and restore approach that dynamically adapts the restoration process. Also, the tight integration of IQA and IR into a unified model minimizes the latency and information loss that typically arises from keeping the two modules disjoint, (e.g. during image and/or text decoding). Extensive experiments show that our approach consistent improvements under single, unknown and composite degradations, thereby establishing a new state-of-the-art.

Yassine Ouali, Adrian Bulat, Brais Martinez et al.

Despite recent successes, LVLMs or Large Vision Language Models are prone to hallucinating details like objects and their properties or relations, limiting their real-world deployment. To address this and improve their robustness, we present CLIP-DPO, a preference optimization method that leverages contrastively pre-trained Vision-Language (VL) embedding models, such as CLIP, for DPO-based optimization of LVLMs. Unlike prior works tackling LVLM hallucinations, our method does not rely on paid-for APIs, and does not require additional training data or the deployment of other external LVLMs. Instead, starting from the initial pool of supervised fine-tuning data, we generate a diverse set of predictions, which are ranked based on their CLIP image-text similarities, and then filtered using a robust rule-based approach to obtain a set of positive and negative pairs for DPO-based training. We applied CLIP-DPO fine-tuning to the MobileVLM-v2 family of models and to LlaVA-1.5, in all cases observing significant improvements in terms of hallucination reduction over baseline models. We also observe better performance for zero-shot classification, suggesting improved grounding capabilities, and verify that the original performance on standard LVLM benchmarks is overall preserved.

Yassine Ouali, Adrian Bulat, Brais Martinez et al.

Vision-Language (V-L) models trained with contrastive learning to align the visual and language modalities have been shown to be strong few-shot learners. Soft prompt learning is the method of choice for few-shot downstream adaptation aiming to bridge the modality gap caused by the distribution shift induced by the new domain. While parameter-efficient, prompt learning still requires access to the model weights and can be computationally infeasible for large models with billions of parameters. To address these shortcomings, in this work, we describe a black-box method for V-L few-shot adaptation that (a) operates on pre-computed image and text features and hence works without access to the model's weights, (b) it is orders of magnitude faster at training time, (c) it is amenable to both supervised and unsupervised training, and (d) it can be even used to align image and text features computed from uni-modal models. To achieve this, we propose Linear Feature Alignment (LFA), a simple linear approach for V-L re-alignment in the target domain. LFA is initialized from a closed-form solution to a least-squares problem and then it is iteratively updated by minimizing a re-ranking loss. Despite its simplicity, our approach can even surpass soft-prompt learning methods as shown by extensive experiments on 11 image and 2 video datasets.

Zhonglin Sun, Georgios Tzimiropoulos

Holistic methods using CNNs and margin-based losses have dominated research on face recognition. In this work, we depart from this setting in two ways: (a) we employ the Vision Transformer as an architecture for training a very strong baseline for face recognition, simply called fViT, which already surpasses most state-of-the-art face recognition methods. (b) Secondly, we capitalize on the Transformer's inherent property to process information (visual tokens) extracted from irregular grids to devise a pipeline for face recognition which is reminiscent of part-based face recognition methods. Our pipeline, called part fViT, simply comprises a lightweight network to predict the coordinates of facial landmarks followed by the Vision Transformer operating on patches extracted from the predicted landmarks, and it is trained end-to-end with no landmark supervision. By learning to extract discriminative patches, our part-based Transformer further boosts the accuracy of our Vision Transformer baseline achieving state-of-the-art accuracy on several face recognition benchmarks.

Ioannis Maniadis Metaxas, Adrian Bulat, Ioannis Patras et al.

The unsupervised pretraining of object detectors has recently become a key component of object detector training, as it leads to improved performance and faster convergence during the supervised fine-tuning stage. Existing unsupervised pretraining methods, however, typically rely on low-level information to define proposals that are used to train the detector. Furthermore, in the absence of class labels for these proposals, an auxiliary loss is used to add high-level semantics. This results in complex pipelines and a task gap between the pretraining and the downstream task. We propose a framework that mitigates this issue and consists of three simple yet key ingredients: (i) richer initial proposals that do encode high-level semantics, (ii) class pseudo-labeling through clustering, that enables pretraining using a standard object detection training pipeline, (iii) self-training to iteratively improve and enrich the object proposals. Once the pretraining and downstream tasks are aligned, a simple detection pipeline without further bells and whistles can be directly used for pretraining and, in fact, results in state-of-the-art performance on both the full and low data regimes, across detector architectures and datasets, by significant margins. We further show that our pretraining strategy is also capable of pretraining from scratch (including the backbone) and works on complex images like COCO, paving the path for unsupervised representation learning using object detection directly as a pretext task.

Fuwen Tan, Royson Lee, Łukasz Dudziak et al.

Large language models (LLMs) have revolutionized language processing, delivering outstanding results across multiple applications. However, deploying LLMs on edge devices poses several challenges with respect to memory, energy, and compute costs, limiting their widespread use in devices such as mobile phones. A promising solution is to reduce the number of bits used to represent weights and activations. While existing works have found partial success at quantizing LLMs to lower bitwidths, e.g. 4-bit weights, quantizing activations beyond 16 bits often leads to large computational overheads due to poor on-device quantization support, or a considerable accuracy drop. Yet, 8-bit activations are very attractive for on-device deployment as they would enable LLMs to fully exploit mobile-friendly hardware, e.g. Neural Processing Units (NPUs). In this work, we make a first attempt to facilitate the on-device deployment of LLMs using integer-only quantization. We first investigate the limitations of existing quantization methods for on-device deployment, with a special focus on activation quantization. We then address these limitations by introducing a simple post-training quantization method, named MobileQuant, that extends previous weight equivalent transformation works by jointly optimizing the weight transformation and activation range parameters in an end-to-end manner. MobileQuant demonstrates superior capabilities over existing methods by 1) achieving near-lossless quantization on a wide range of LLM benchmarks, 2) reducing latency and energy consumption by 20\%-50\% compared to current on-device quantization strategies, 3) requiring limited compute budget, 4) being compatible with mobile-friendly compute units, e.g. NPU.

Chen Feng, Georgios Tzimiropoulos, Ioannis Patras

Learning with Noisy labels (LNL) poses a significant challenge for the Machine Learning community. Some of the most widely used approaches that select as clean samples for which the model itself (the in-training model) has high confidence, e.g., `small loss', can suffer from the so called `self-confirmation' bias. This bias arises because the in-training model, is at least partially trained on the noisy labels. Furthermore, in the classification case, an additional challenge arises because some of the label noise is between classes that are visually very similar (`hard noise'). This paper addresses these challenges by proposing a method (\textit{CLIPCleaner}) that leverages CLIP, a powerful Vision-Language (VL) model for constructing a zero-shot classifier for efficient, offline, clean sample selection. This has the advantage that the sample selection is decoupled from the in-training model and that the sample selection is aware of the semantic and visual similarities between the classes due to the way that CLIP is trained. We provide theoretical justifications and empirical evidence to demonstrate the advantages of CLIP for LNL compared to conventional pre-trained models. Compared to current methods that combine iterative sample selection with various techniques, \textit{CLIPCleaner} offers a simple, single-step approach that achieves competitive or superior performance on benchmark datasets. To the best of our knowledge, this is the first time a VL model has been used for sample selection to address the problem of Learning with Noisy Labels (LNL), highlighting their potential in the domain.

Adrian Bulat, Enrique Sanchez, Brais Martinez et al.

This work is on training a generative action/video recognition model whose output is a free-form action-specific caption describing the video (rather than an action class label). A generative approach has practical advantages like producing more fine-grained and human-readable output, and being naturally open-world. To this end, we propose to adapt a pre-trained generative Vision & Language (V&L) Foundation Model for video/action recognition. While recently there have been a few attempts to adapt V&L models trained with contrastive learning (e.g. CLIP) for video/action, to the best of our knowledge, we propose the very first method that sets outs to accomplish this goal for a generative model. We firstly show that direct fine-tuning of a generative model to produce action classes suffers from severe overfitting. To alleviate this, we introduce REST, a training framework consisting of two key components: an unsupervised method for adapting the generative model to action/video by means of pseudo-caption generation and Self-training, i.e. without using any action-specific labels; (b) a Retrieval approach based on CLIP for discovering a diverse set of pseudo-captions for each video to train the model. Importantly, we show that both components are necessary to obtain high accuracy. We evaluate REST on the problem of zero-shot action recognition where we show that our approach is very competitive when compared to contrastive learning-based methods. Code will be made available.

Zhonglin Sun, Siyang Song, Ioannis Patras et al.

Privacy issue is a main concern in developing face recognition techniques. Although synthetic face images can partially mitigate potential legal risks while maintaining effective face recognition (FR) performance, FR models trained by face images synthesized by existing generative approaches frequently suffer from performance degradation problems due to the insufficient discriminative quality of these synthesized samples. In this paper, we systematically investigate what contributes to solid face recognition model training, and reveal that face images with certain degree of similarities to their identity centers show great effectiveness in the performance of trained FR models. Inspired by this, we propose a novel diffusion-based approach (namely Center-based Semi-hard Synthetic Face Generation (CemiFace)) which produces facial samples with various levels of similarity to the subject center, thus allowing to generate face datasets containing effective discriminative samples for training face recognition. Experimental results show that with a modest degree of similarity, training on the generated dataset can produce competitive performance compared to previous generation methods.

Dimitrios Mallis, Enrique Sanchez, Matt Bell et al.

This paper proposes a novel paradigm for the unsupervised learning of object landmark detectors. Contrary to existing methods that build on auxiliary tasks such as image generation or equivariance, we propose a self-training approach where, departing from generic keypoints, a landmark detector and descriptor is trained to improve itself, tuning the keypoints into distinctive landmarks. To this end, we propose an iterative algorithm that alternates between producing new pseudo-labels through feature clustering and learning distinctive features for each pseudo-class through contrastive learning. With a shared backbone for the landmark detector and descriptor, the keypoint locations progressively converge to stable landmarks, filtering those less stable. Compared to previous works, our approach can learn points that are more flexible in terms of capturing large viewpoint changes. We validate our method on a variety of difficult datasets, including LS3D, BBCPose, Human3.6M and PennAction, achieving new state of the art results.

Adrian Bulat, Brais Martinez, Georgios Tzimiropoulos

This paper tackles the problem of efficient video recognition. In this area, video transformers have recently dominated the efficiency (top-1 accuracy vs FLOPs) spectrum. At the same time, there have been some attempts in the image domain which challenge the necessity of the self-attention operation within the transformer architecture, advocating the use of simpler approaches for token mixing. However, there are no results yet for the case of video recognition, where the self-attention operator has a significantly higher impact (compared to the case of images) on efficiency. To address this gap, in this paper, we make the following contributions: (a) we construct a highly efficient \& accurate attention-free block based on the shift operator, coined Affine-Shift block, specifically designed to approximate as closely as possible the operations in the MHSA block of a Transformer layer. Based on our Affine-Shift block, we construct our Affine-Shift Transformer and show that it already outperforms all existing shift/MLP--based architectures for ImageNet classification. (b) We extend our formulation in the video domain to construct Video Affine-Shift Transformer (VAST), the very first purely attention-free shift-based video transformer. (c) We show that VAST significantly outperforms recent state-of-the-art transformers on the most popular action recognition benchmarks for the case of models with low computational and memory footprint. Code will be made available.

Fatemeh Saleh, Fuwen Tan, Adrian Bulat et al.

Learning visual representations through self-supervision is an extremely challenging task as the network needs to sieve relevant patterns from spurious distractors without the active guidance provided by supervision. This is achieved through heavy data augmentation, large-scale datasets and prohibitive amounts of compute. Video self-supervised learning (SSL) suffers from added challenges: video datasets are typically not as large as image datasets, compute is an order of magnitude larger, and the amount of spurious patterns the optimizer has to sieve through is multiplied several fold. Thus, directly learning self-supervised representations from video data might result in sub-optimal performance. To address this, we propose to utilize a strong image-based model, pre-trained with self- or language supervision, in a video representation learning framework, enabling the model to learn strong spatial and temporal information without relying on the video labeled data. To this end, we modify the typical video-based SSL design and objective to encourage the video encoder to \textit{subsume} the semantic content of an image-based model trained on a general domain. The proposed algorithm is shown to learn much more efficiently (i.e. in less epochs and with a smaller batch) and results in a new state-of-the-art performance on standard downstream tasks among single-modality SSL methods.

Adrian Bulat, Alberto Baldrati, Ioannis Maniadis Metaxas et al.

Existing approaches for improving the efficiency of Large Vision-Language Models (LVLMs) are largely based on the concept of visual token reduction. This approach, however, creates an information bottleneck that impairs performance, especially on challenging tasks that require fine-grained understanding and reasoning. In this work, we challenge this paradigm by introducing VISion On Request (VISOR), a method that reduces inference cost without discarding visual information. Instead of compressing the image, VISOR improves efficiency by sparsifying the interaction between image and text tokens. Specifically, the language model attends to the full set of high-resolution visual tokens through a small, strategically placed set of attention layers: general visual context is provided by efficient cross-attention between text-image, while a few well-placed and dynamically selected self-attention layers refine the visual representations themselves, enabling complex, high-resolution reasoning when needed. Based on this principle, we first train a single universal network on a range of computational budgets by varying the number of self-attention layers, and then introduce a lightweight policy mechanism that dynamically allocates visual computation based on per-sample complexity. Extensive experiments show that VISOR drastically reduces computational cost while matching or exceeding state-of-the-art results across a diverse suite of benchmarks, and excels in challenging tasks that require detailed visual understanding.

Ioannis Maniadis Metaxas, Georgios Tzimiropoulos, Ioannis Patras

Self-supervised learning has recently emerged as the preeminent pretraining paradigm across and between modalities, with remarkable results. In the image domain specifically, group (or cluster) discrimination has been one of the most successful methods. However, such frameworks need to guard against heavily imbalanced cluster assignments to prevent collapse to trivial solutions. Existing works typically solve this by reweighing cluster assignments to promote balance, or with offline operations (e.g. regular re-clustering) that prevent collapse. However, the former typically requires large batch sizes, which leads to increased resource requirements, and the latter introduces scalability issues with regard to large datasets. In this work, we propose ExCB, a framework that tackles this problem with a novel cluster balancing method. ExCB estimates the relative size of the clusters across batches and balances them by adjusting cluster assignments, proportionately to their relative size and in an online manner. Thereby, it overcomes previous methods' dependence on large batch sizes and is fully online, and therefore scalable to any dataset. We conduct extensive experiments to evaluate our approach and demonstrate that ExCB: a) achieves state-of-the-art results with significantly reduced resource requirements compared to previous works, b) is fully online, and therefore scalable to large datasets, and c) is stable and effective even with very small batch sizes.

Marios Krestenitis, Christos Tzelepis, Konstantinos Ioannidis et al.

Visual-Language Models (VLMs) have achieved remarkable progress in image captioning, visual question answering, and visual reasoning. Yet they remain prone to vision-language misalignment, often producing overly generic or hallucinated descriptions. Existing approaches address this via instruction tuning-requiring costly, large-scale annotated datasets or via complex test-time frameworks for caption refinement. In this work, we revisit image-text alignment through the lens of cycle consistency: given an image and a caption generated by an image-to-text model, the backward mapping through a text-to-image model should reconstruct an image that closely matches the original. In our setup, a VLM serves as the image-to-text component, while a pre-trained text-to-image model closes the loop by reconstructing the image from the generated caption. Building on this, we introduce CycleCap, a fine-tuning scheme to improve image captioning using Group Relative Policy Optimization (GRPO) with a reward based on the similarity between the original and reconstructed images, computed on-the-fly. Unlike previous work that uses cycle consistency loss for preference dataset construction, our method leverages cycle consistency directly as a self-supervised training signal. This enables the use of raw images alone, eliminating the need for curated image-text datasets, while steering the VLM to produce more accurate and grounded text descriptions. Applied to four VLMs ranging from 1B to 7B parameters, CycleCap yields consistent improvements across captioning and hallucination benchmarks, surpassing state-of-the-art methods that rely on supervised cycle consistency training.

Alexandros Xenos, Niki Maria Foteinopoulou, Ioanna Ntinou et al.

Recognising emotions in context involves identifying an individual's apparent emotions while considering contextual cues from the surrounding scene. Previous approaches to this task have typically designed explicit scene-encoding architectures or incorporated external scene-related information, such as captions. However, these methods often utilise limited contextual information or rely on intricate training pipelines to decouple noise from relevant information. In this work, we leverage the capabilities of Vision-and-Large-Language Models (VLLMs) to enhance in-context emotion classification in a more straightforward manner. Our proposed method follows a simple yet effective two-stage approach. First, we prompt VLLMs to generate natural language descriptions of the subject's apparent emotion in relation to the visual context. Second, the descriptions, along with the visual input, are used to train a transformer-based architecture that fuses text and visual features before the final classification task. This method not only simplifies the training process but also significantly improves performance. Experimental results demonstrate that the textual descriptions effectively guide the model to constrain the noisy visual input, allowing our fused architecture to outperform individual modalities. Our approach achieves state-of-the-art performance across three datasets, BoLD, EMOTIC, and CAER-S, without bells and whistles. The code will be made publicly available on github: https://github.com/NickyFot/EmoCommonSense.git

Ioanna Ntinou, Enrique Sanchez, Georgios Tzimiropoulos

Action Localization is a challenging problem that combines detection and recognition tasks, which are often addressed separately. State-of-the-art methods rely on off-the-shelf bounding box detections pre-computed at high resolution, and propose transformer models that focus on the classification task alone. Such two-stage solutions are prohibitive for real-time deployment. On the other hand, single-stage methods target both tasks by devoting part of the network (generally the backbone) to sharing the majority of the workload, compromising performance for speed. These methods build on adding a DETR head with learnable queries that after cross- and self-attention can be sent to corresponding MLPs for detecting a person's bounding box and action. However, DETR-like architectures are challenging to train and can incur in big complexity. In this paper, we observe that \textbf{a straight bipartite matching loss can be applied to the output tokens of a vision transformer}. This results in a backbone + MLP architecture that can do both tasks without the need of an extra encoder-decoder head and learnable queries. We show that a single MViTv2-S architecture trained with bipartite matching to perform both tasks surpasses the same MViTv2-S when trained with RoI align on pre-computed bounding boxes. With a careful design of token pooling and the proposed training pipeline, our Bipartite-Matching Vision Transformer model, \textbf{BMViT}, achieves +3 mAP on AVA2.2. w.r.t. the two-stage MViTv2-S counterpart. Code is available at \href{https://github.com/IoannaNti/BMViT}{https://github.com/IoannaNti/BMViT}

Isma Hadji, Enrique Sanchez, Adrian Bulat et al.

We introduce a multi-scale Image Super Resolution (ISR) method building on recent advances in Visual Auto-Regressive (VAR) modeling. VAR models break image tokenization into additive, gradually increasing scales, using Residual Quantization (RQ), an approach that aligns perfectly with our target ISR task. Previous works taking advantage of this synergy suffer from two main shortcomings. First, due to the limitations in RQ, they only generate images at a predefined fixed scale, failing to map intermediate outputs to the corresponding image scales. They also rely on large backbones or a large corpus of annotated data to achieve better performance. To address both shortcomings, we introduce two novel components to the VAR training for ISR, aiming at increasing its flexibility and reducing its complexity. In particular, we introduce a) a \textbf{Hierarchical Image Tokenization (HIT)} approach that progressively represents images at different scales while enforcing token overlap across scales, and b) a \textbf{Direct Preference Optimization (DPO) regularization term} that, relying solely on the (LR,HR) pair, encourages the transformer to produce the latter over the former. Our proposed HIT acts as a strong inductive bias for the VAR training, resulting in a small model (300M params vs 1B params of VARSR), that achieves state-of-the-art results without external training data, and that delivers multi-scale outputs with a single forward pass.

Ioanna Ntinou, Alexandros Xenos, Yassine Ouali et al.

Contrastively-trained Vision-Language Models (VLMs), such as CLIP, have become the standard approach for learning discriminative vision-language representations. However, these models often exhibit shallow language understanding, manifesting bag-of-words behaviour. These limitations are reinforced by their dual-encoder design, which induces a modality gap. Additionally, the reliance on vast web-collected data corpora for training makes the process computationally expensive and introduces significant privacy concerns. To address these limitations, in this work, we challenge the necessity of vision encoders for retrieval tasks by introducing a vision-free, single-encoder retrieval pipeline. Departing from the traditional text-to-image retrieval paradigm, we migrate to a text-to-text paradigm with the assistance of VLLM-generated structured image descriptions. We demonstrate that this paradigm shift has significant advantages, including a substantial reduction of the modality gap, improved compositionality, and better performance on short and long caption queries, all attainable with only a few hours of calibration on two GPUs. Additionally, substituting raw images with textual descriptions introduces a more privacy-friendly alternative for retrieval. To further assess generalisation and address some of the shortcomings of prior compositionality benchmarks, we release two benchmarks derived from Flickr30k and COCO, containing diverse compositional queries made of short captions, which we coin subFlickr and subCOCO. Our vision-free retriever matches and often surpasses traditional multimodal models. Importantly, our approach achieves state-of-the-art zero-shot performance on multiple retrieval and compositionality benchmarks, with models as small as 0.3B parameters. Code is available at: https://github.com/IoannaNti/LexiCLIP

Stella Bounareli, Vasileios Argyriou, Georgios Tzimiropoulos

This paper is on face/head reenactment where the goal is to transfer the facial pose (3D head orientation and expression) of a target face to a source face. Previous methods focus on learning embedding networks for identity and pose disentanglement which proves to be a rather hard task, degrading the quality of the generated images. We take a different approach, bypassing the training of such networks, by using (fine-tuned) pre-trained GANs which have been shown capable of producing high-quality facial images. Because GANs are characterized by weak controllability, the core of our approach is a method to discover which directions in latent GAN space are responsible for controlling facial pose and expression variations. We present a simple pipeline to learn such directions with the aid of a 3D shape model which, by construction, already captures disentangled directions for facial pose, identity and expression. Moreover, we show that by embedding real images in the GAN latent space, our method can be successfully used for the reenactment of real-world faces. Our method features several favorable properties including using a single source image (one-shot) and enabling cross-person reenactment. Our qualitative and quantitative results show that our approach often produces reenacted faces of significantly higher quality than those produced by state-of-the-art methods for the standard benchmarks of VoxCeleb1 & 2. Source code is available at: https://github.com/StelaBou/stylegan_directions_face_reenactment

Chen Feng, Georgios Tzimiropoulos, Ioannis Patras

Despite the large progress in supervised learning with neural networks, there are significant challenges in obtaining high-quality, large-scale and accurately labelled datasets. In such a context, how to learn in the presence of noisy labels has received more and more attention. As a relatively complex problem, in order to achieve good results, current approaches often integrate components from several fields, such as supervised learning, semi-supervised learning, transfer learning and resulting in complicated methods. Furthermore, they often make multiple assumptions about the type of noise of the data. This affects the model robustness and limits its performance under different noise conditions. In this paper, we consider a novel problem setting, Learning with Unknown Label Noise}(LULN), that is, learning when both the degree and the type of noise are unknown. Under this setting, unlike previous methods that often introduce multiple assumptions and lead to complex solutions, we propose a simple, efficient and robust framework named Sample Selection and Relabelling(SSR), that with a minimal number of hyperparameters achieves SOTA results in various conditions. At the heart of our method is a sample selection and relabelling mechanism based on a non-parametric KNN classifier~(NPK) $g_q$ and a parametric model classifier~(PMC) $g_p$, respectively, to select the clean samples and gradually relabel the noisy samples. Without bells and whistles, such as model co-training, self-supervised pre-training and semi-supervised learning, and with robustness concerning the settings of its few hyper-parameters, our method significantly surpasses previous methods on both CIFAR10/CIFAR100 with synthetic noise and real-world noisy datasets such as WebVision, Clothing1M and ANIMAL-10N. Code is available at https://github.com/MrChenFeng/SSR_BMVC2022.

Christos Tzelepis, Georgios Tzimiropoulos, Ioannis Patras

This work addresses the problem of discovering, in an unsupervised manner, interpretable paths in the latent space of pretrained GANs, so as to provide an intuitive and easy way of controlling the underlying generative factors. In doing so, it addresses some of the limitations of the state-of-the-art works, namely, a) that they discover directions that are independent of the latent code, i.e., paths that are linear, and b) that their evaluation relies either on visual inspection or on laborious human labeling. More specifically, we propose to learn non-linear warpings on the latent space, each one parametrized by a set of RBF-based latent space warping functions, and where each warping gives rise to a family of non-linear paths via the gradient of the function. Building on the work of Voynov and Babenko, that discovers linear paths, we optimize the trainable parameters of the set of RBFs, so as that images that are generated by codes along different paths, are easily distinguishable by a discriminator network. This leads to easily distinguishable image transformations, such as pose and facial expressions in facial images. We show that linear paths can be derived as a special case of our method, and show experimentally that non-linear paths in the latent space lead to steeper, more disentangled and interpretable changes in the image space than in state-of-the art methods, both qualitatively and quantitatively. We make the code and the pretrained models publicly available at: https://github.com/chi0tzp/WarpedGANSpace.

Brais Martinez, Jing Yang, Adrian Bulat et al.

This paper shows how to train binary networks to within a few percent points ($\sim 3-5 \%$) of the full precision counterpart. We first show how to build a strong baseline, which already achieves state-of-the-art accuracy, by combining recently proposed advances and carefully adjusting the optimization procedure. Secondly, we show that by attempting to minimize the discrepancy between the output of the binary and the corresponding real-valued convolution, additional significant accuracy gains can be obtained. We materialize this idea in two complementary ways: (1) with a loss function, during training, by matching the spatial attention maps computed at the output of the binary and real-valued convolutions, and (2) in a data-driven manner, by using the real-valued activations, available during inference prior to the binarization process, for re-scaling the activations right after the binary convolution. Finally, we show that, when putting all of our improvements together, the proposed model beats the current state of the art by more than 5% top-1 accuracy on ImageNet and reduces the gap to its real-valued counterpart to less than 3% and 5% top-1 accuracy on CIFAR-100 and ImageNet respectively when using a ResNet-18 architecture. Code available at https://github.com/brais-martinez/real2binary.

Adrian Bulat, Brais Martinez, Georgios Tzimiropoulos

This paper proposes Binary ArchitecTure Search (BATS), a framework that drastically reduces the accuracy gap between binary neural networks and their real-valued counterparts by means of Neural Architecture Search (NAS). We show that directly applying NAS to the binary domain provides very poor results. To alleviate this, we describe, to our knowledge, for the first time, the 3 key ingredients for successfully applying NAS to the binary domain. Specifically, we (1) introduce and design a novel binary-oriented search space, (2) propose a new mechanism for controlling and stabilising the resulting searched topologies, (3) propose and validate a series of new search strategies for binary networks that lead to faster convergence and lower search times. Experimental results demonstrate the effectiveness of the proposed approach and the necessity of searching in the binary space directly. Moreover, (4) we set a new state-of-the-art for binary neural networks on CIFAR10, CIFAR100 and ImageNet datasets. Code will be made available https://github.com/1adrianb/binary-nas

Enrique Sanchez, Georgios Tzimiropoulos

This paper proposes a method to ease the unsupervised learning of object landmark detectors. Similarly to previous methods, our approach is fully unsupervised in a sense that it does not require or make any use of annotated landmarks for the target object category. Contrary to previous works, we do however assume that a landmark detector, which has already learned a structured representation for a given object category in a fully supervised manner, is available. Under this setting, our main idea boils down to adapting the given pre-trained network to the target object categories in a fully unsupervised manner. To this end, our method uses the pre-trained network as a core which remains frozen and does not get updated during training, and learns, in an unsupervised manner, only a projection matrix to perform the adaptation to the target categories. By building upon an existing structured representation learned in a supervised manner, the optimization problem solved by our method is much more constrained with significantly less parameters to learn which seems to be important for the case of unsupervised learning. We show that our method surpasses fully unsupervised techniques trained from scratch as well as a strong baseline based on fine-tuning, and produces state-of-the-art results on several datasets. Code can be found at https://github.com/ESanchezLozano/SAIC-Unsupervised-landmark-detection-NeurIPS2019 .

Enrique Sanchez-Lozano, Georgios Tzimiropoulos, Michel Valstar

This paper proposes a supervised learning approach to jointly perform facial Action Unit (AU) localisation and intensity estimation. Contrary to previous works that try to learn an unsupervised representation of the Action Unit regions, we propose to directly and jointly estimate all AU intensities through heatmap regression, along with the location in the face where they cause visible changes. Our approach aims to learn a pixel-wise regression function returning a score per AU, which indicates an AU intensity at a given spatial location. Heatmap regression then generates an image, or channel, per AU, in which each pixel indicates the corresponding AU intensity. To generate the ground-truth heatmaps for a target AU, the facial landmarks are first estimated, and a 2D Gaussian is drawn around the points where the AU is known to cause changes. The amplitude and size of the Gaussian is determined by the intensity of the AU. We show that using a single Hourglass network suffices to attain new state of the art results, demonstrating the effectiveness of such a simple approach. The use of heatmap regression allows learning of a shared representation between AUs without the need to rely on latent representations, as these are implicitly learned from the data. We validate the proposed approach on the BP4D dataset, showing a modest improvement on recent, complex, techniques, as well as robustness against misalignment errors. Code for testing and models will be available to download from https://github.com/ESanchezLozano/Action-Units-Heatmaps.

Thi Minh Anh Pham, An Duc Nguyen, Cephas Svosve et al.

Large pre-trained vision-language models such as CLIP have demonstrated great potential in zero-shot transferability to downstream tasks. However, to attain optimal performance, the manual selection of prompts is necessary to improve alignment between the downstream image distribution and the textual class descriptions. This manual prompt engineering is the major challenge for deploying such models in practice since it requires domain expertise and is extremely time-consuming. To avoid non-trivial prompt engineering, recent work Context Optimization (CoOp) introduced the concept of prompt learning to the vision domain using learnable textual tokens. While CoOp can achieve substantial improvements over manual prompts, its learned context is worse generalizable to wider unseen classes within the same dataset. In this work, we present Prompt Learning with Reparameterization Encoder (PRE) - a simple and efficient method that enhances the generalization ability of the learnable prompt to unseen classes while maintaining the capacity to learn Base classes. Instead of directly optimizing the prompts, PRE employs a prompt encoder to reparameterize the input prompt embeddings, enhancing the exploration of task-specific knowledge from few-shot samples. Experiments and extensive ablation studies on 8 benchmarks demonstrate that our approach is an efficient method for prompt learning. Specifically, PRE achieves a notable enhancement of 5.60% in average accuracy on New classes and 3% in Harmonic mean compared to CoOp in the 16-shot setting, all achieved within a good training time.

Mehdi Noroozi, Isma Hadji, Brais Martinez et al.

In this paper, we introduce YONOS-SR, a novel stable diffusion-based approach for image super-resolution that yields state-of-the-art results using only a single DDIM step. We propose a novel scale distillation approach to train our SR model. Instead of directly training our SR model on the scale factor of interest, we start by training a teacher model on a smaller magnification scale, thereby making the SR problem simpler for the teacher. We then train a student model for a higher magnification scale, using the predictions of the teacher as a target during the training. This process is repeated iteratively until we reach the target scale factor of the final model. The rationale behind our scale distillation is that the teacher aids the student diffusion model training by i) providing a target adapted to the current noise level rather than using the same target coming from ground truth data for all noise levels and ii) providing an accurate target as the teacher has a simpler task to solve. We empirically show that the distilled model significantly outperforms the model trained for high scales directly, specifically with few steps during inference. Having a strong diffusion model that requires only one step allows us to freeze the U-Net and fine-tune the decoder on top of it. We show that the combination of spatially distilled U-Net and fine-tuned decoder outperforms state-of-the-art methods requiring 200 steps with only one single step.

Zhonglin Sun, Chen Feng, Ioannis Patras et al.

In this work we focus on learning facial representations that can be adapted to train effective face recognition models, particularly in the absence of labels. Firstly, compared with existing labelled face datasets, a vastly larger magnitude of unlabeled faces exists in the real world. We explore the learning strategy of these unlabeled facial images through self-supervised pretraining to transfer generalized face recognition performance. Moreover, motivated by one recent finding, that is, the face saliency area is critical for face recognition, in contrast to utilizing random cropped blocks of images for constructing augmentations in pretraining, we utilize patches localized by extracted facial landmarks. This enables our method - namely LAndmark-based Facial Self-supervised learning LAFS), to learn key representation that is more critical for face recognition. We also incorporate two landmark-specific augmentations which introduce more diversity of landmark information to further regularize the learning. With learned landmark-based facial representations, we further adapt the representation for face recognition with regularization mitigating variations in landmark positions. Our method achieves significant improvement over the state-of-the-art on multiple face recognition benchmarks, especially on more challenging few-shot scenarios.

Stella Bounareli, Christos Tzelepis, Vasileios Argyriou et al.

In this paper, we present our framework for neural face/head reenactment whose goal is to transfer the 3D head orientation and expression of a target face to a source face. Previous methods focus on learning embedding networks for identity and head pose/expression disentanglement which proves to be a rather hard task, degrading the quality of the generated images. We take a different approach, bypassing the training of such networks, by using (fine-tuned) pre-trained GANs which have been shown capable of producing high-quality facial images. Because GANs are characterized by weak controllability, the core of our approach is a method to discover which directions in latent GAN space are responsible for controlling head pose and expression variations. We present a simple pipeline to learn such directions with the aid of a 3D shape model which, by construction, inherently captures disentangled directions for head pose, identity, and expression. Moreover, we show that by embedding real images in the GAN latent space, our method can be successfully used for the reenactment of real-world faces. Our method features several favorable properties including using a single source image (one-shot) and enabling cross-person reenactment. Extensive qualitative and quantitative results show that our approach typically produces reenacted faces of notably higher quality than those produced by state-of-the-art methods for the standard benchmarks of VoxCeleb1 & 2.

Yassine Ouali, Adrian Bulat, Alexandros Xenos et al.

Contrastively-trained Vision-Language Models (VLMs) like CLIP have become the de facto approach for discriminative vision-language representation learning. However, these models have limited language understanding, often exhibiting a "bag of words" behavior. At the same time, Large Vision-Language Models (LVLMs), which combine vision encoders with LLMs, have been shown to be capable of detailed vision-language reasoning, yet their autoregressive nature renders them less suitable for discriminative tasks. In this work, we propose to combine "the best of both worlds": a new training approach for discriminative fine-tuning of LVLMs that results in strong discriminative and compositional capabilities. Essentially, our approach converts a generative LVLM into a discriminative one, unlocking its capability for powerful image-text discrimination combined with enhanced language understanding. Our contributions include (1) a carefully designed training/optimization framework that utilizes image-text pairs of variable length and granularity for training the model with both contrastive and next-token prediction losses. This is accompanied by ablation studies that justify the necessity of our framework's components; (2) a parameter-efficient adaptation method using a combination of soft prompting and LoRA adapters; (3) significant improvements over state-of-the-art CLIP-like models of similar size, including standard image-text retrieval benchmarks and notable gains in compositionality.

Stella Bounareli, Christos Tzelepis, Vasileios Argyriou et al.

Video-driven neural face reenactment aims to synthesize realistic facial images that successfully preserve the identity and appearance of a source face, while transferring the target head pose and facial expressions. Existing GAN-based methods suffer from either distortions and visual artifacts or poor reconstruction quality, i.e., the background and several important appearance details, such as hair style/color, glasses and accessories, are not faithfully reconstructed. Recent advances in Diffusion Probabilistic Models (DPMs) enable the generation of high-quality realistic images. To this end, in this paper we present DiffusionAct, a novel method that leverages the photo-realistic image generation of diffusion models to perform neural face reenactment. Specifically, we propose to control the semantic space of a Diffusion Autoencoder (DiffAE), in order to edit the facial pose of the input images, defined as the head pose orientation and the facial expressions. Our method allows one-shot, self, and cross-subject reenactment, without requiring subject-specific fine-tuning. We compare against state-of-the-art GAN-, StyleGAN2-, and diffusion-based methods, showing better or on-par reenactment performance.

Adrian Bulat, Yassine Ouali, Georgios Tzimiropoulos

Despite noise and caption quality having been acknowledged as important factors impacting vision-language contrastive pre-training, in this paper, we show that the full potential of improving the training process by addressing such issues is yet to be realized. Specifically, we firstly study and analyze two issues affecting training: incorrect assignment of negative pairs, and low caption quality and diversity. Then, we devise effective solutions for addressing both problems, which essentially require training with multiple true positive pairs. Finally, we propose training with sigmoid loss to address such a requirement. We show very large gains over the current state-of-the-art for both image recognition ($\sim +6\%$ on average over 11 datasets) and image retrieval ($\sim +19\%$ on Flickr30k and $\sim +15\%$ on MSCOCO).

Mehdi Noroozi, Isma Hadji, Victor Escorcia et al.

There has been immense progress recently in the visual quality of Stable Diffusion-based Super Resolution (SD-SR). However, deploying large diffusion models on computationally restricted devices such as mobile phones remains impractical due to the large model size and high latency. This is compounded for SR as it often operates at high res (e.g. 4Kx3K). In this work, we introduce Edge-SD-SR, the first parameter efficient and low latency diffusion model for image super-resolution. Edge-SD-SR consists of ~169M parameters, including UNet, encoder and decoder, and has a complexity of only ~142 GFLOPs. To maintain a high visual quality on such low compute budget, we introduce a number of training strategies: (i) A novel conditioning mechanism on the low resolution input, coined bidirectional conditioning, which tailors the SD model for the SR task. (ii) Joint training of the UNet and encoder, while decoupling the encodings of the HR and LR images and using a dedicated schedule. (iii) Finetuning the decoder using the UNet's output to directly tailor the decoder to the latents obtained at inference time. Edge-SD-SR runs efficiently on device, e.g. it can upscale a 128x128 patch to 512x512 in 38 msec while running on a Samsung S24 DSP, and of a 512x512 to 2048x2048 (requiring 25 model evaluations) in just ~1.1 sec. Furthermore, we show that Edge-SD-SR matches or even outperforms state-of-the-art SR approaches on the most established SR benchmarks.

Haosen Yang, Adrian Bulat, Isma Hadji et al.

Diffusion models are proficient at generating high-quality images. They are however effective only when operating at the resolution used during training. Inference at a scaled resolution leads to repetitive patterns and structural distortions. Retraining at higher resolutions quickly becomes prohibitive. Thus, methods enabling pre-existing diffusion models to operate at flexible test-time resolutions are highly desirable. Previous works suffer from frequent artifacts and often introduce large latency overheads. We propose two simple modules that combine to solve these issues. We introduce a Frequency Modulation (FM) module that leverages the Fourier domain to improve the global structure consistency, and an Attention Modulation (AM) module which improves the consistency of local texture patterns, a problem largely ignored in prior works. Our method, coined Fam diffusion, can seamlessly integrate into any latent diffusion model and requires no additional training. Extensive qualitative results highlight the effectiveness of our method in addressing structural and local artifacts, while quantitative results show state-of-the-art performance. Also, our method avoids redundant inference tricks for improved consistency such as patch-based or progressive generation, leading to negligible latency overheads.

Chen Feng, Zhuo Zhi, Zhao Huang et al.

Statistically consistent methods based on the noise transition matrix ($T$) offer a theoretically grounded solution to Learning with Noisy Labels (LNL), with guarantees of convergence to the optimal clean-data classifier. In practice, however, these methods are often outperformed by empirical approaches such as sample selection, and this gap is usually attributed to the difficulty of accurately estimating $T$. The common assumption is that, given a perfect $T$, noise-correction methods would recover their theoretical advantage. In this work, we put this longstanding hypothesis to a decisive test. We conduct experiments under idealized conditions, providing correction methods with a perfect, oracle transition matrix. Even under these ideal conditions, we observe that these methods still suffer from performance collapse during training. This compellingly demonstrates that the failure is not fundamentally a $T$-estimation problem, but stems from a more deeply rooted flaw. To explain this behaviour, we provide a unified analysis that links three levels: macroscopic convergence states, microscopic optimisation dynamics, and information-theoretic limits on what can be learned from noisy labels. Together, these results give a formal account of why ideal noise correction fails and offer concrete guidance for designing more reliable methods for learning with noisy labels.

Debin Meng, Chen Jin, Zheng Gao et al.

Image diversity remains a fundamental challenge for text-to-image diffusion models. Low-diversity models tend to generate repetitive outputs, increasing sampling redundancy and hindering both creative exploration and downstream applications. A primary cause is that generation often collapses toward a strong mode in the learned distribution. Existing attempts to improve diversity, such as noise resampling, prompt rewriting, or steering-based guidance, often still collapse to dominant modes or introduce distortions that degrade image quality. In light of this, we propose Token-Prompt embedding Space Optimization (TPSO), a training-free and model-agnostic module. TPSO introduces learnable parameters to explore underrepresented regions of the token embedding space, reducing the tendency of the model to repeatedly generate samples from strong modes of the learned distribution. At the same time, the prompt-level space provides a global semantic constraint that regulates distribution shifts, preventing quality degradation while maintaining high fidelity. Extensive experiments on MS-COCO and three diffusion backbones show that TPSO significantly enhances generative diversity, improving baseline performance from 1.10 to 4.18 points, without sacrificing image quality. Code will be released upon acceptance.

Enrique Sanchez, Isma Hadji, Adrian Bulat et al.

In this paper we tackle Image Super Resolution (ISR), using recent advances in Visual Auto-Regressive (VAR) modeling. VAR iteratively estimates the residual in latent space between gradually increasing image scales, a process referred to as next-scale prediction. Thus, the strong priors learned during pre-training align well with the downstream task (ISR). To our knowledge, only VARSR has exploited this synergy so far, showing promising results. However, due to the limitations of existing residual quantizers, VARSR works only at a fixed resolution, i.e. it fails to map intermediate outputs to the corresponding image scales. Additionally, it relies on a 1B transformer architecture (VAR-d24), and leverages a large-scale private dataset to achieve state-of-the-art results. We address these limitations through two novel components: a) a Hierarchical Image Tokenization approach with a multi-scale image tokenizer that progressively represents images at different scales while simultaneously enforcing token overlap across scales, and b) a Direct Preference Optimization (DPO) regularization term that, relying solely on the LR and HR tokenizations, encourages the transformer to produce the latter over the former. To the best of our knowledge, this is the first time a quantizer is trained to force semantically consistent residuals at different scales, and the first time that preference-based optimization is used to train a VAR. Using these two components, our model can denoise the LR image and super-resolve at half and full target upscale factors in a single forward pass. Additionally, we achieve \textit{state-of-the-art results on ISR}, while using a small model (300M params vs ~1B params of VARSR), and without using external training data.