Philip H. S. Torr

Shuyang Sun, Weijun Wang, Qihang Yu et al.

This paper presents a new mechanism to facilitate the training of mask transformers for efficient panoptic segmentation, democratizing its deployment. We observe that due to its high complexity, the training objective of panoptic segmentation will inevitably lead to much higher false positive penalization. Such unbalanced loss makes the training process of the end-to-end mask-transformer based architectures difficult, especially for efficient models. In this paper, we present ReMaX that adds relaxation to mask predictions and class predictions during training for panoptic segmentation. We demonstrate that via these simple relaxation techniques during training, our model can be consistently improved by a clear margin \textbf{without} any extra computational cost on inference. By combining our method with efficient backbones like MobileNetV3-Small, our method achieves new state-of-the-art results for efficient panoptic segmentation on COCO, ADE20K and Cityscapes. Code and pre-trained checkpoints will be available at \url{https://github.com/google-research/deeplab2}.

Zeyu Yang, Nan Song, Wei Li et al.

Existing top-performance autonomous driving systems typically rely on the multi-modal fusion strategy for reliable scene understanding. This design is however fundamentally restricted due to overlooking the modality-specific strengths and finally hampering the model performance. To address this limitation, in this work, we introduce a novel modality interaction strategy that allows individual per-modality representations to be learned and maintained throughout, enabling their unique characteristics to be exploited during the whole perception pipeline. To demonstrate the effectiveness of the proposed strategy, we design DeepInteraction++, a multi-modal interaction framework characterized by a multi-modal representational interaction encoder and a multi-modal predictive interaction decoder. Specifically, the encoder is implemented as a dual-stream Transformer with specialized attention operation for information exchange and integration between separate modality-specific representations. Our multi-modal representational learning incorporates both object-centric, precise sampling-based feature alignment and global dense information spreading, essential for the more challenging planning task. The decoder is designed to iteratively refine the predictions by alternately aggregating information from separate representations in a unified modality-agnostic manner, realizing multi-modal predictive interaction. Extensive experiments demonstrate the superior performance of the proposed framework on both 3D object detection and end-to-end autonomous driving tasks. Our code is available at https://github.com/fudan-zvg/DeepInteraction.

Dubing Chen, Yuming Shen, Haofeng Zhang et al.

Recent research on Generalized Zero-Shot Learning (GZSL) has focused primarily on generation-based methods. However, current literature has overlooked the fundamental principles of these methods and has made limited progress in a complex manner. In this paper, we aim to deconstruct the generator-classifier framework and provide guidance for its improvement and extension. We begin by breaking down the generator-learned unseen class distribution into class-level and instance-level distributions. Through our analysis of the role of these two types of distributions in solving the GZSL problem, we generalize the focus of the generation-based approach, emphasizing the importance of (i) attribute generalization in generator learning and (ii) independent classifier learning with partially biased data. We present a simple method based on this analysis that outperforms SotAs on four public GZSL datasets, demonstrating the validity of our deconstruction. Furthermore, our proposed method remains effective even without a generative model, representing a step towards simplifying the generator-classifier structure. Our code is available at \url{https://github.com/cdb342/DGZ}.

Yoshua Bengio, Geoffrey Hinton, Andrew Yao et al. · mila

Artificial Intelligence (AI) is progressing rapidly, and companies are shifting their focus to developing generalist AI systems that can autonomously act and pursue goals. Increases in capabilities and autonomy may soon massively amplify AI's impact, with risks that include large-scale social harms, malicious uses, and an irreversible loss of human control over autonomous AI systems. Although researchers have warned of extreme risks from AI, there is a lack of consensus about how exactly such risks arise, and how to manage them. Society's response, despite promising first steps, is incommensurate with the possibility of rapid, transformative progress that is expected by many experts. AI safety research is lagging. Present governance initiatives lack the mechanisms and institutions to prevent misuse and recklessness, and barely address autonomous systems. In this short consensus paper, we describe extreme risks from upcoming, advanced AI systems. Drawing on lessons learned from other safety-critical technologies, we then outline a comprehensive plan combining technical research and development with proactive, adaptive governance mechanisms for a more commensurate preparation.

Francisco Eiras, Kemal Oksuz, Adel Bibi et al.

Referring Image Segmentation (RIS) - the problem of identifying objects in images through natural language sentences - is a challenging task currently mostly solved through supervised learning. However, while collecting referred annotation masks is a time-consuming process, the few existing weakly-supervised and zero-shot approaches fall significantly short in performance compared to fully-supervised learning ones. To bridge the performance gap without mask annotations, we propose a novel weakly-supervised framework that tackles RIS by decomposing it into three steps: obtaining instance masks for the object mentioned in the referencing instruction (segment), using zero-shot learning to select a potentially correct mask for the given instruction (select), and bootstrapping a model which allows for fixing the mistakes of zero-shot selection (correct). In our experiments, using only the first two steps (zero-shot segment and select) outperforms other zero-shot baselines by as much as 16.5%, while our full method improves upon this much stronger baseline and sets the new state-of-the-art for weakly-supervised RIS, reducing the gap between the weakly-supervised and fully-supervised methods in some cases from around 33% to as little as 7%. Code is available at https://github.com/fgirbal/segment-select-correct.

Runjia Li, Shuyang Sun, Mohamed Elhoseiny et al. · oxford

This paper presents OxfordTVG-HIC (Humorous Image Captions), a large-scale dataset for humour generation and understanding. Humour is an abstract, subjective, and context-dependent cognitive construct involving several cognitive factors, making it a challenging task to generate and interpret. Hence, humour generation and understanding can serve as a new task for evaluating the ability of deep-learning methods to process abstract and subjective information. Due to the scarcity of data, humour-related generation tasks such as captioning remain under-explored. To address this gap, OxfordTVG-HIC offers approximately 2.9M image-text pairs with humour scores to train a generalizable humour captioning model. Contrary to existing captioning datasets, OxfordTVG-HIC features a wide range of emotional and semantic diversity resulting in out-of-context examples that are particularly conducive to generating humour. Moreover, OxfordTVG-HIC is curated devoid of offensive content. We also show how OxfordTVG-HIC can be leveraged for evaluating the humour of a generated text. Through explainability analysis of the trained models, we identify the visual and linguistic cues influential for evoking humour prediction (and generation). We observe qualitatively that these cues are aligned with the benign violation theory of humour in cognitive psychology.

Kejie Li, Jia-Wang Bian, Robert Castle et al. · bytedance, oxford

High-quality 3D ground-truth shapes are critical for 3D object reconstruction evaluation. However, it is difficult to create a replica of an object in reality, and even 3D reconstructions generated by 3D scanners have artefacts that cause biases in evaluation. To address this issue, we introduce a novel multi-view RGBD dataset captured using a mobile device, which includes highly precise 3D ground-truth annotations for 153 object models featuring a diverse set of 3D structures. We obtain precise 3D ground-truth shape without relying on high-end 3D scanners by utilising LEGO models with known geometry as the 3D structures for image capture. The distinct data modality offered by high-resolution RGB images and low-resolution depth maps captured on a mobile device, when combined with precise 3D geometry annotations, presents a unique opportunity for future research on high-fidelity 3D reconstruction. Furthermore, we evaluate a range of 3D reconstruction algorithms on the proposed dataset. Project page: http://code.active.vision/MobileBrick/

Henghui Ding, Chang Liu, Shuting He et al.

Video object segmentation (VOS) aims at segmenting a particular object throughout the entire video clip sequence. The state-of-the-art VOS methods have achieved excellent performance (e.g., 90+% J&F) on existing datasets. However, since the target objects in these existing datasets are usually relatively salient, dominant, and isolated, VOS under complex scenes has rarely been studied. To revisit VOS and make it more applicable in the real world, we collect a new VOS dataset called coMplex video Object SEgmentation (MOSE) to study the tracking and segmenting objects in complex environments. MOSE contains 2,149 video clips and 5,200 objects from 36 categories, with 431,725 high-quality object segmentation masks. The most notable feature of MOSE dataset is complex scenes with crowded and occluded objects. The target objects in the videos are commonly occluded by others and disappear in some frames. To analyze the proposed MOSE dataset, we benchmark 18 existing VOS methods under 4 different settings on the proposed MOSE dataset and conduct comprehensive comparisons. The experiments show that current VOS algorithms cannot well perceive objects in complex scenes. For example, under the semi-supervised VOS setting, the highest J&F by existing state-of-the-art VOS methods is only 59.4% on MOSE, much lower than their ~90% J&F performance on DAVIS. The results reveal that although excellent performance has been achieved on existing benchmarks, there are unresolved challenges under complex scenes and more efforts are desired to explore these challenges in the future. The proposed MOSE dataset has been released at https://henghuiding.github.io/MOSE.

Li Zhang, Mohan Chen, Anurag Arnab et al.

Modelling long-range dependencies is critical for scene understanding tasks in computer vision. Although convolution neural networks (CNNs) have excelled in many vision tasks, they are still limited in capturing long-range structured relationships as they typically consist of layers of local kernels. A fully-connected graph, such as the self-attention operation in Transformers, is beneficial for such modelling, however, its computational overhead is prohibitive. In this paper, we propose a dynamic graph message passing network, that significantly reduces the computational complexity compared to related works modelling a fully-connected graph. This is achieved by adaptively sampling nodes in the graph, conditioned on the input, for message passing. Based on the sampled nodes, we dynamically predict node-dependent filter weights and the affinity matrix for propagating information between them. This formulation allows us to design a self-attention module, and more importantly a new Transformer-based backbone network, that we use for both image classification pretraining, and for addressing various downstream tasks (object detection, instance and semantic segmentation). Using this model, we show significant improvements with respect to strong, state-of-the-art baselines on four different tasks. Our approach also outperforms fully-connected graphs while using substantially fewer floating-point operations and parameters. Code and models will be made publicly available at https://github.com/fudan-zvg/DGMN2

Weiming Hu, Qiang Wang, Li Zhang et al.

In this paper we introduce SiamMask, a framework to perform both visual object tracking and video object segmentation, in real-time, with the same simple method. We improve the offline training procedure of popular fully-convolutional Siamese approaches by augmenting their losses with a binary segmentation task. Once the offline training is completed, SiamMask only requires a single bounding box for initialization and can simultaneously carry out visual object tracking and segmentation at high frame-rates. Moreover, we show that it is possible to extend the framework to handle multiple object tracking and segmentation by simply re-using the multi-task model in a cascaded fashion. Experimental results show that our approach has high processing efficiency, at around 55 frames per second. It yields real-time state-of-the-art results on visual-object tracking benchmarks, while at the same time demonstrating competitive performance at a high speed for video object segmentation benchmarks.

Jishnu Mukhoti, Yarin Gal, Philip H. S. Torr et al.

Pre-trained foundation models, due to their enormous capacity and exposure to vast amounts of data during pre-training, are known to have learned plenty of real-world concepts. An important step in making these pre-trained models effective on downstream tasks is to fine-tune them on related datasets. While various fine-tuning methods have been devised and have been shown to be highly effective, we observe that a fine-tuned model's ability to recognize concepts on tasks $\textit{different}$ from the downstream one is reduced significantly compared to its pre-trained counterpart. This is an undesirable effect of fine-tuning as a substantial amount of resources was used to learn these pre-trained concepts in the first place. We call this phenomenon ''concept forgetting'' and via experiments show that most end-to-end fine-tuning approaches suffer heavily from this side effect. To this end, we propose a simple fix to this problem by designing a new fine-tuning method called $\textit{LDIFS}$ (short for $\ell_2$ distance in feature space) that, while learning new concepts related to the downstream task, allows a model to preserve its pre-trained knowledge as well. Through extensive experiments on 10 fine-tuning tasks we show that $\textit{LDIFS}$ significantly reduces concept forgetting. Additionally, we show that LDIFS is highly effective in performing continual fine-tuning on a sequence of tasks as well, in comparison with both fine-tuning as well as continual learning baselines.

Yasir Ghunaim, Adel Bibi, Kumail Alhamoud et al.

Current evaluations of Continual Learning (CL) methods typically assume that there is no constraint on training time and computation. This is an unrealistic assumption for any real-world setting, which motivates us to propose: a practical real-time evaluation of continual learning, in which the stream does not wait for the model to complete training before revealing the next data for predictions. To do this, we evaluate current CL methods with respect to their computational costs. We conduct extensive experiments on CLOC, a large-scale dataset containing 39 million time-stamped images with geolocation labels. We show that a simple baseline outperforms state-of-the-art CL methods under this evaluation, questioning the applicability of existing methods in realistic settings. In addition, we explore various CL components commonly used in the literature, including memory sampling strategies and regularization approaches. We find that all considered methods fail to be competitive against our simple baseline. This surprisingly suggests that the majority of existing CL literature is tailored to a specific class of streams that is not practical. We hope that the evaluation we provide will be the first step towards a paradigm shift to consider the computational cost in the development of online continual learning methods.

Nan Xue, Tianfu Wu, Song Bai et al.

This article presents Holistically-Attracted Wireframe Parsing (HAWP), a method for geometric analysis of 2D images containing wireframes formed by line segments and junctions. HAWP utilizes a parsimonious Holistic Attraction (HAT) field representation that encodes line segments using a closed-form 4D geometric vector field. The proposed HAWP consists of three sequential components empowered by end-to-end and HAT-driven designs: (1) generating a dense set of line segments from HAT fields and endpoint proposals from heatmaps, (2) binding the dense line segments to sparse endpoint proposals to produce initial wireframes, and (3) filtering false positive proposals through a novel endpoint-decoupled line-of-interest aligning (EPD LOIAlign) module that captures the co-occurrence between endpoint proposals and HAT fields for better verification. Thanks to our novel designs, HAWPv2 shows strong performance in fully supervised learning, while HAWPv3 excels in self-supervised learning, achieving superior repeatability scores and efficient training (24 GPU hours on a single GPU). Furthermore, HAWPv3 exhibits a promising potential for wireframe parsing in out-of-distribution images without providing ground truth labels of wireframes.

Francesco Pinto, Harry Yang, Ser-Nam Lim et al.

We show that the effectiveness of the well celebrated Mixup [Zhang et al., 2018] can be further improved if instead of using it as the sole learning objective, it is utilized as an additional regularizer to the standard cross-entropy loss. This simple change not only provides much improved accuracy but also significantly improves the quality of the predictive uncertainty estimation of Mixup in most cases under various forms of covariate shifts and out-of-distribution detection experiments. In fact, we observe that Mixup yields much degraded performance on detecting out-of-distribution samples possibly, as we show empirically, because of its tendency to learn models that exhibit high-entropy throughout; making it difficult to differentiate in-distribution samples from out-distribution ones. To show the efficacy of our approach (RegMixup), we provide thorough analyses and experiments on vision datasets (ImageNet & CIFAR-10/100) and compare it with a suite of recent approaches for reliable uncertainty estimation.

Aleksandar Petrov, Philip H. S. Torr, Adel Bibi

Context-based fine-tuning methods, including prompting, in-context learning, soft prompting (also known as prompt tuning), and prefix-tuning, have gained popularity due to their ability to often match the performance of full fine-tuning with a fraction of the parameters. Despite their empirical successes, there is little theoretical understanding of how these techniques influence the internal computation of the model and their expressiveness limitations. We show that despite the continuous embedding space being more expressive than the discrete token space, soft-prompting and prefix-tuning are potentially less expressive than full fine-tuning, even with the same number of learnable parameters. Concretely, context-based fine-tuning cannot change the relative attention pattern over the content and can only bias the outputs of an attention layer in a fixed direction. This suggests that while techniques like prompting, in-context learning, soft prompting, and prefix-tuning can effectively elicit skills present in the pretrained model, they may not be able to learn novel tasks that require new attention patterns.

Jia-Wang Bian, Wenjing Bian, Victor Adrian Prisacariu et al. · bytedance, oxford

Neural surface reconstruction is sensitive to the camera pose noise, even if state-of-the-art pose estimators like COLMAP or ARKit are used. More importantly, existing Pose-NeRF joint optimisation methods have struggled to improve pose accuracy in challenging real-world scenarios. To overcome the challenges, we introduce the pose residual field (PoRF), a novel implicit representation that uses an MLP for regressing pose updates. This is more robust than the conventional pose parameter optimisation due to parameter sharing that leverages global information over the entire sequence. Furthermore, we propose an epipolar geometry loss to enhance the supervision that leverages the correspondences exported from COLMAP results without the extra computational overhead. Our method yields promising results. On the DTU dataset, we reduce the rotation error by 78\% for COLMAP poses, leading to the decreased reconstruction Chamfer distance from 3.48mm to 0.85mm. On the MobileBrick dataset that contains casually captured unbounded 360-degree videos, our method refines ARKit poses and improves the reconstruction F1 score from 69.18 to 75.67, outperforming that with the dataset provided ground-truth pose (75.14). These achievements demonstrate the efficacy of our approach in refining camera poses and improving the accuracy of neural surface reconstruction in real-world scenarios.

Tim Franzmeyer, Philip H. S. Torr, João F. Henriques

We study how an autonomous agent learns to perform a task from demonstrations in a different domain, such as a different environment or different agent. Such cross-domain imitation learning is required to, for example, train an artificial agent from demonstrations of a human expert. We propose a scalable framework that enables cross-domain imitation learning without access to additional demonstrations or further domain knowledge. We jointly train the learner agent's policy and learn a mapping between the learner and expert domains with adversarial training. We effect this by using a mutual information criterion to find an embedding of the expert's state space that contains task-relevant information and is invariant to domain specifics. This step significantly simplifies estimating the mapping between the learner and expert domains and hence facilitates end-to-end learning. We demonstrate successful transfer of policies between considerably different domains, without extra supervision such as additional demonstrations, and in situations where other methods fail.

Jindong Gu, Fangyun Wei, Philip Torr et al. · deepmind, oxford

Deep Neural Networks can be easily fooled by small and imperceptible perturbations. The query-based black-box attack (QBBA) is able to create the perturbations using model output probabilities of image queries requiring no access to the underlying models. QBBA poses realistic threats to real-world applications. Recently, various types of robustness have been explored to defend against QBBA. In this work, we first taxonomize the stochastic defense strategies against QBBA. Following our taxonomy, we propose to explore non-additive randomness in models to defend against QBBA. Specifically, we focus on underexplored Vision Transformers based on their flexible architectures. Extensive experiments show that the proposed defense approach achieves effective defense, without much sacrifice in performance.

Xipeng Chen, Guangrun Wang, Dizhong Zhu et al.

3D Garment modeling is a critical and challenging topic in the area of computer vision and graphics, with increasing attention focused on garment representation learning, garment reconstruction, and controllable garment manipulation, whereas existing methods were constrained to model garments under specific categories or with relatively simple topologies. In this paper, we propose a novel Neural Sewing Machine (NSM), a learning-based framework for structure-preserving 3D garment modeling, which is capable of learning representations for garments with diverse shapes and topologies and is successfully applied to 3D garment reconstruction and controllable manipulation. To model generic garments, we first obtain sewing pattern embedding via a unified sewing pattern encoding module, as the sewing pattern can accurately describe the intrinsic structure and the topology of the 3D garment. Then we use a 3D garment decoder to decode the sewing pattern embedding into a 3D garment using the UV-position maps with masks. To preserve the intrinsic structure of the predicted 3D garment, we introduce an inner-panel structure-preserving loss, an inter-panel structure-preserving loss, and a surface-normal loss in the learning process of our framework. We evaluate NSM on the public 3D garment dataset with sewing patterns with diverse garment shapes and categories. Extensive experiments demonstrate that the proposed NSM is capable of representing 3D garments under diverse garment shapes and topologies, realistically reconstructing 3D garments from 2D images with the preserved structure, and accurately manipulating the 3D garment categories, shapes, and topologies, outperforming the state-of-the-art methods by a clear margin.

Bowen Li, Philip H. S. Torr, Thomas Lukasiewicz

We introduce a memory-driven semi-parametric approach to text-to-image generation, which is based on both parametric and non-parametric techniques. The non-parametric component is a memory bank of image features constructed from a training set of images. The parametric component is a generative adversarial network. Given a new text description at inference time, the memory bank is used to selectively retrieve image features that are provided as basic information of target images, which enables the generator to produce realistic synthetic results. We also incorporate the content information into the discriminator, together with semantic features, allowing the discriminator to make a more reliable prediction. Experimental results demonstrate that the proposed memory-driven semi-parametric approach produces more realistic images than purely parametric approaches, in terms of both visual fidelity and text-image semantic consistency.

Guangrun Wang, Philip H. S. Torr

Classifiers and generators have long been separated. We break down this separation and showcase that conventional neural network classifiers can generate high-quality images of a large number of categories, being comparable to the state-of-the-art generative models (e.g., DDPMs and GANs). We achieve this by computing the partial derivative of the classification loss function with respect to the input to optimize the input to produce an image. Since it is widely known that directly optimizing the inputs is similar to targeted adversarial attacks incapable of generating human-meaningful images, we propose a mask-based stochastic reconstruction module to make the gradients semantic-aware to synthesize plausible images. We further propose a progressive-resolution technique to guarantee fidelity, which produces photorealistic images. Furthermore, we introduce a distance metric loss and a non-trivial distribution loss to ensure classification neural networks can synthesize diverse and high-fidelity images. Using traditional neural network classifiers, we can generate good-quality images of 256$\times$256 resolution on ImageNet. Intriguingly, our method is also applicable to text-to-image generation by regarding image-text foundation models as generalized classifiers. Proving that classifiers have learned the data distribution and are ready for image generation has far-reaching implications, for classifiers are much easier to train than generative models like DDPMs and GANs. We don't even need to train classification models because tons of public ones are available for download. Also, this holds great potential for the interpretability and robustness of classifiers. Project page is at \url{https://classifier-as-generator.github.io/}.

Jishnu Mukhoti, Tsung-Yu Lin, Bor-Chun Chen et al.

In image classification, a lot of development has happened in detecting out-of-distribution (OoD) data. However, most OoD detection methods are evaluated on a standard set of datasets, arbitrarily different from training data. There is no clear definition of what forms a ``good" OoD dataset. Furthermore, the state-of-the-art OoD detection methods already achieve near perfect results on these standard benchmarks. In this paper, we define 2 categories of OoD data using the subtly different concepts of perceptual/visual and semantic similarity to in-distribution (iD) data. We define Near OoD samples as perceptually similar but semantically different from iD samples, and Shifted samples as points which are visually different but semantically akin to iD data. We then propose a GAN based framework for generating OoD samples from each of these 2 categories, given an iD dataset. Through extensive experiments on MNIST, CIFAR-10/100 and ImageNet, we show that a) state-of-the-art OoD detection methods which perform exceedingly well on conventional benchmarks are significantly less robust to our proposed benchmark. Moreover, b) models performing well on our setup also perform well on conventional real-world OoD detection benchmarks and vice versa, thereby indicating that one might not even need a separate OoD set, to reliably evaluate performance in OoD detection.

Li Zhang, Jiachen Lu, Sixiao Zheng et al.

The emergence of vision transformers (ViTs) in image classification has shifted the methodologies for visual representation learning. In particular, ViTs learn visual representation at full receptive field per layer across all the image patches, in comparison to the increasing receptive fields of CNNs across layers and other alternatives (e.g., large kernels and atrous convolution). In this work, for the first time we explore the global context learning potentials of ViTs for dense visual prediction (e.g., semantic segmentation). Our motivation is that through learning global context at full receptive field layer by layer, ViTs may capture stronger long-range dependency information, critical for dense prediction tasks. We first demonstrate that encoding an image as a sequence of patches, a vanilla ViT without local convolution and resolution reduction can yield stronger visual representation for semantic segmentation. For example, our model, termed as SEgmentation TRansformer (SETR), excels on ADE20K (50.28% mIoU, the first position in the test leaderboard on the day of submission) and performs competitively on Cityscapes. However, the basic ViT architecture falls short in broader dense prediction applications, such as object detection and instance segmentation, due to its lack of a pyramidal structure, high computational demand, and insufficient local context. For tackling general dense visual prediction tasks in a cost-effective manner, we further formulate a family of Hierarchical Local-Global (HLG) Transformers, characterized by local attention within windows and global-attention across windows in a pyramidal architecture. Extensive experiments show that our methods achieve appealing performance on a variety of dense prediction tasks (e.g., object detection and instance segmentation and semantic segmentation) as well as image classification.

Francesco Pinto, Nathalie Rauschmayr, Florian Tramèr et al.

Vision-Language Models (VLMs) have made remarkable progress in document-based Visual Question Answering (i.e., responding to queries about the contents of an input document provided as an image). In this work, we show these models can memorize responses for training samples and regurgitate them even when the relevant visual information has been removed. This includes Personal Identifiable Information (PII) repeated once in the training set, indicating these models could divulge memorised sensitive information and therefore pose a privacy risk. We quantitatively measure the extractability of information in controlled experiments and differentiate between cases where it arises from generalization capabilities or from memorization. We further investigate the factors that influence memorization across multiple state-of-the-art models and propose an effective heuristic countermeasure that empirically prevents the extractability of PII.

Guangyi Chen, Xiao Liu, Guangrun Wang et al.

Video-language pre-trained models have shown remarkable success in guiding video question-answering (VideoQA) tasks. However, due to the length of video sequences, training large-scale video-based models incurs considerably higher costs than training image-based ones. This motivates us to leverage the knowledge from image-based pretraining, despite the obvious gaps between image and video domains. To bridge these gaps, in this paper, we propose Tem-Adapter, which enables the learning of temporal dynamics and complex semantics by a visual Temporal Aligner and a textual Semantic Aligner. Unlike conventional pretrained knowledge adaptation methods that only concentrate on the downstream task objective, the Temporal Aligner introduces an extra language-guided autoregressive task aimed at facilitating the learning of temporal dependencies, with the objective of predicting future states based on historical clues and language guidance that describes event progression. Besides, to reduce the semantic gap and adapt the textual representation for better event description, we introduce a Semantic Aligner that first designs a template to fuse question and answer pairs as event descriptions and then learns a Transformer decoder with the whole video sequence as guidance for refinement. We evaluate Tem-Adapter and different pre-train transferring methods on two VideoQA benchmarks, and the significant performance improvement demonstrates the effectiveness of our method.

Ekaterina Iakovleva, Fabio Pizzati, Philip Torr et al.

Text-based editing diffusion models exhibit limited performance when the user's input instruction is ambiguous. To solve this problem, we propose $\textit{Specify ANd Edit}$ (SANE), a zero-shot inference pipeline for diffusion-based editing systems. We use a large language model (LLM) to decompose the input instruction into specific instructions, i.e. well-defined interventions to apply to the input image to satisfy the user's request. We benefit from the LLM-derived instructions along the original one, thanks to a novel denoising guidance strategy specifically designed for the task. Our experiments with three baselines and on two datasets demonstrate the benefits of SANE in all setups. Moreover, our pipeline improves the interpretability of editing models, and boosts the output diversity. We also demonstrate that our approach can be applied to any edit, whether ambiguous or not. Our code is public at https://github.com/fabvio/SANE.

Runjia Li, Junlin Han, Luke Melas-Kyriazi et al. · oxford

We present DreamBeast, a novel method based on score distillation sampling (SDS) for generating fantastical 3D animal assets composed of distinct parts. Existing SDS methods often struggle with this generation task due to a limited understanding of part-level semantics in text-to-image diffusion models. While recent diffusion models, such as Stable Diffusion 3, demonstrate a better part-level understanding, they are prohibitively slow and exhibit other common problems associated with single-view diffusion models. DreamBeast overcomes this limitation through a novel part-aware knowledge transfer mechanism. For each generated asset, we efficiently extract part-level knowledge from the Stable Diffusion 3 model into a 3D Part-Affinity implicit representation. This enables us to instantly generate Part-Affinity maps from arbitrary camera views, which we then use to modulate the guidance of a multi-view diffusion model during SDS to create 3D assets of fantastical animals. DreamBeast significantly enhances the quality of generated 3D creatures with user-specified part compositions while reducing computational overhead, as demonstrated by extensive quantitative and qualitative evaluations.

Tim Franzmeyer, Stephen McAleer, João F. Henriques et al.

Autonomous agents deployed in the real world need to be robust against adversarial attacks on sensory inputs. Robustifying agent policies requires anticipating the strongest attacks possible. We demonstrate that existing observation-space attacks on reinforcement learning agents have a common weakness: while effective, their lack of information-theoretic detectability constraints makes them detectable using automated means or human inspection. Detectability is undesirable to adversaries as it may trigger security escalations. We introduce ε-illusory, a novel form of adversarial attack on sequential decision-makers that is both effective and of ε-bounded statistical detectability. We propose a novel dual ascent algorithm to learn such attacks end-to-end. Compared to existing attacks, we empirically find ε-illusory to be significantly harder to detect with automated methods, and a small study with human participants (IRB approval under reference R84123/RE001) suggests they are similarly harder to detect for humans. Our findings suggest the need for better anomaly detectors, as well as effective hardware- and system-level defenses. The project website can be found at https://tinyurl.com/illusory-attacks.

Alex Dobra, Jakiw Pidstrigach, Tim Reichelt et al.

Sensitivity analysis is a cornerstone of climate science, essential for understanding phenomena ranging from storm intensity to long-term climate feedbacks. However, computing these sensitivities using traditional physical models is often prohibitively expensive in terms of both computation and development time. While modern AI-based generative models are orders of magnitude faster to evaluate, computing sensitivities with them remains a significant bottleneck. This work addresses this challenge by applying the adjoint state method for calculating gradients in generative flow models. We apply this method to the cBottle generative model, trained on ERA5 and ICON data, to perform sensitivity analysis of any atmospheric variable with respect to sea surface temperatures. We quantitatively validate the computed sensitivities against the model's own outputs. Our results provide initial evidence that this approach can produce reliable gradients, reducing the computational cost of sensitivity analysis from weeks on a supercomputer with a physical model to hours on a GPU, thereby simplifying a critical workflow in climate science. The code can be found at https://github.com/Kwartzl8/cbottle_adjoint_sensitivity.

Kumail Alhamoud, Yasir Ghunaim, Motasem Alfarra et al.

For medical imaging AI models to be clinically impactful, they must generalize. However, this goal is hindered by (i) diverse types of distribution shifts, such as temporal, demographic, and label shifts, and (ii) limited diversity in datasets that are siloed within single medical institutions. While these limitations have spurred interest in federated learning, current evaluation benchmarks fail to evaluate different shifts simultaneously. However, in real healthcare settings, multiple types of shifts co-exist, yet their impact on medical imaging performance remains unstudied. In response, we introduce FedMedICL, a unified framework and benchmark to holistically evaluate federated medical imaging challenges, simultaneously capturing label, demographic, and temporal distribution shifts. We comprehensively evaluate several popular methods on six diverse medical imaging datasets (totaling 550 GPU hours). Furthermore, we use FedMedICL to simulate COVID-19 propagation across hospitals and evaluate whether methods can adapt to pandemic changes in disease prevalence. We find that a simple batch balancing technique surpasses advanced methods in average performance across FedMedICL experiments. This finding questions the applicability of results from previous, narrow benchmarks in real-world medical settings.

Hasan Abed Al Kader Hammoud, Hani Itani, Fabio Pizzati et al.

We present SynthCLIP, a CLIP model trained on entirely synthetic text-image pairs. Leveraging recent text-to-image (TTI) networks and large language models (LLM), we generate synthetic datasets of images and corresponding captions at scale, with no human intervention. In this work, we provide an analysis on CLIP models trained on synthetic data. We provide insights on the data generation strategy, number of samples required, scaling trends, and resulting properties. We also introduce SynthCI-30M, a purely synthetic dataset comprising 30 million captioned images. Our code, trained models, and data, are released as open source at https://github.com/hammoudhasan/SynthCLIP

Andrew M. Bean, Ryan Othniel Kearns, Angelika Romanou et al.

Evaluating large language models (LLMs) is crucial for both assessing their capabilities and identifying safety or robustness issues prior to deployment. Reliably measuring abstract and complex phenomena such as 'safety' and 'robustness' requires strong construct validity, that is, having measures that represent what matters to the phenomenon. With a team of 29 expert reviewers, we conduct a systematic review of 445 LLM benchmarks from leading conferences in natural language processing and machine learning. Across the reviewed articles, we find patterns related to the measured phenomena, tasks, and scoring metrics which undermine the validity of the resulting claims. To address these shortcomings, we provide eight key recommendations and detailed actionable guidance to researchers and practitioners in developing LLM benchmarks.

Yuanze Lin, Yunsheng Li, Dongdong Chen et al.

In recent years, multimodal large language models (MLLMs) have made significant strides by training on vast high-quality image-text datasets, enabling them to generally understand images well. However, the inherent difficulty in explicitly conveying fine-grained or spatially dense information in text, such as masks, poses a challenge for MLLMs, limiting their ability to answer questions requiring an understanding of detailed or localized visual elements. Drawing inspiration from the Retrieval-Augmented Generation (RAG) concept, this paper proposes a new visual prompt approach to integrate fine-grained external knowledge, gleaned from specialized vision models (e.g., instance segmentation/OCR models), into MLLMs. This is a promising yet underexplored direction for enhancing MLLMs' performance. Our approach diverges from concurrent works, which transform external knowledge into additional text prompts, necessitating the model to indirectly learn the correspondence between visual content and text coordinates. Instead, we propose embedding fine-grained knowledge information directly into a spatial embedding map as a visual prompt. This design can be effortlessly incorporated into various MLLMs, such as LLaVA and Mipha, considerably improving their visual understanding performance. Through rigorous experiments, we demonstrate that our method can enhance MLLM performance across nine benchmarks, amplifying their fine-grained context-aware capabilities.

Lama Alssum, Hani Itani, Hasan Abed Al Kader Hammoud et al.

The safety alignment of large language models (LLMs) is becoming increasingly important with their democratization. In this paper, we study the safety degradation that comes with adapting LLMs to new tasks. We attribute this safety compromise to catastrophic forgetting and frame the problem of preserving safety when fine-tuning as a continual learning (CL) problem. We consider the fine-tuning-as-a-service setup where the user uploads their data to a service provider to get a customized model that excels on the user's selected task. We adapt several CL approaches from the literature and systematically evaluate their ability to mitigate safety degradation. These include regularization-based, memory-based, and model merging approaches. We consider two scenarios, (1) benign user data and (2) poisoned user data. Our results demonstrate that CL approaches consistently achieve lower attack success rates than standard fine-tuning. Among these, DER outperforms both other CL methods and existing safety-preserving baselines while maintaining task utility. These findings generalize across three downstream tasks (GSM8K, SST2, Code) and three model families (LLaMA2-7B, Mistral-7B, Gemma-2B), establishing CL as a practical solution to preserve safety.

Hasan Abed Al Kader Hammoud, Tuhin Das, Fabio Pizzati et al.

We explore the impact of training with more diverse datasets, characterized by the number of unique samples, on the performance of self-supervised learning (SSL) under a fixed computational budget. Our findings consistently demonstrate that increasing pretraining data diversity enhances SSL performance, albeit only when the distribution distance to the downstream data is minimal. Notably, even with an exceptionally large pretraining data diversity achieved through methods like web crawling or diffusion-generated data, among other ways, the distribution shift remains a challenge. Our experiments are comprehensive with seven SSL methods using large-scale datasets such as ImageNet and YFCC100M amounting to over 200 GPU days. Code and trained models are available at https://github.com/hammoudhasan/DiversitySSL

Ameya Prabhu, Shiven Sinha, Ponnurangam Kumaraguru et al.

Continual learning has primarily focused on the issue of catastrophic forgetting and the associated stability-plasticity tradeoffs. However, little attention has been paid to the efficacy of continually learned representations, as representations are learned alongside classifiers throughout the learning process. Our primary contribution is empirically demonstrating that existing online continually trained deep networks produce inferior representations compared to a simple pre-defined random transforms. Our approach projects raw pixels using a fixed random transform, approximating an RBF-Kernel initialized before any data is seen. We then train a simple linear classifier on top without storing any exemplars, processing one sample at a time in an online continual learning setting. This method, called RanDumb, significantly outperforms state-of-the-art continually learned representations across all standard online continual learning benchmarks. Our study reveals the significant limitations of representation learning, particularly in low-exemplar and online continual learning scenarios. Extending our investigation to popular exemplar-free scenarios with pretrained models, we find that training only a linear classifier on top of pretrained representations surpasses most continual fine-tuning and prompt-tuning strategies. Overall, our investigation challenges the prevailing assumptions about effective representation learning in online continual learning. Our code is available at://github.com/drimpossible/RanDumb.

Ling Yang, Zixiang Zhang, Junlin Han et al. · oxford

Generating high-quality 3D assets from textual descriptions remains a pivotal challenge in computer graphics and vision research. Due to the scarcity of 3D data, state-of-the-art approaches utilize pre-trained 2D diffusion priors, optimized through Score Distillation Sampling (SDS). Despite progress, crafting complex 3D scenes featuring multiple objects or intricate interactions is still difficult. To tackle this, recent methods have incorporated box or layout guidance. However, these layout-guided compositional methods often struggle to provide fine-grained control, as they are generally coarse and lack expressiveness. To overcome these challenges, we introduce a novel SDS approach, Semantic Score Distillation Sampling (SemanticSDS), designed to effectively improve the expressiveness and accuracy of compositional text-to-3D generation. Our approach integrates new semantic embeddings that maintain consistency across different rendering views and clearly differentiate between various objects and parts. These embeddings are transformed into a semantic map, which directs a region-specific SDS process, enabling precise optimization and compositional generation. By leveraging explicit semantic guidance, our method unlocks the compositional capabilities of existing pre-trained diffusion models, thereby achieving superior quality in 3D content generation, particularly for complex objects and scenes. Experimental results demonstrate that our SemanticSDS framework is highly effective for generating state-of-the-art complex 3D content. Code: https://github.com/YangLing0818/SemanticSDS-3D

Aleksandar Petrov, Shruti Agarwal, Philip H. S. Torr et al.

Watermarking, the practice of embedding imperceptible information into media such as images, videos, audio, and text, is essential for intellectual property protection, content provenance and attribution. The growing complexity of digital ecosystems necessitates watermarks for different uses to be embedded in the same media. However, to detect and decode all watermarks, they need to coexist well with one another. We perform the first study of coexistence of deep image watermarking methods and, contrary to intuition, we find that various open-source watermarks can coexist with only minor impacts on image quality and decoding robustness. The coexistence of watermarks also opens the avenue for ensembling watermarking methods. We show how ensembling can increase the overall message capacity and enable new trade-offs between capacity, accuracy, robustness and image quality, without needing to retrain the base models.

Shuang Wu, Youtian Lin, Feihu Zhang et al.

Generating high-quality 3D assets from text and images has long been challenging, primarily due to the absence of scalable 3D representations capable of capturing intricate geometry distributions. In this work, we introduce Direct3D, a native 3D generative model scalable to in-the-wild input images, without requiring a multiview diffusion model or SDS optimization. Our approach comprises two primary components: a Direct 3D Variational Auto-Encoder (D3D-VAE) and a Direct 3D Diffusion Transformer (D3D-DiT). D3D-VAE efficiently encodes high-resolution 3D shapes into a compact and continuous latent triplane space. Notably, our method directly supervises the decoded geometry using a semi-continuous surface sampling strategy, diverging from previous methods relying on rendered images as supervision signals. D3D-DiT models the distribution of encoded 3D latents and is specifically designed to fuse positional information from the three feature maps of the triplane latent, enabling a native 3D generative model scalable to large-scale 3D datasets. Additionally, we introduce an innovative image-to-3D generation pipeline incorporating semantic and pixel-level image conditions, allowing the model to produce 3D shapes consistent with the provided conditional image input. Extensive experiments demonstrate the superiority of our large-scale pre-trained Direct3D over previous image-to-3D approaches, achieving significantly better generation quality and generalization ability, thus establishing a new state-of-the-art for 3D content creation. Project page: https://nju-3dv.github.io/projects/Direct3D/.

Tajamul Ashraf, Umair Nawaz, Abdelrahman M. Shaker et al.

Vision language models (VLMs) are increasingly deployed as controllers with access to external tools for complex reasoning and decision-making, yet their effectiveness remains limited by the scarcity of high-quality multimodal trajectories and the cost of manual annotation. We address this challenge with a vision-centric agent tuning framework that automatically synthesizes multimodal trajectories, generates step-wise preference pairs, and trains a VLM controller for robust tool-use reasoning. Our pipeline first constructs M-TRACE, a large-scale dataset of 28.5K multimodal tasks with 177K verified trajectories, enabling imitation-based trajectory tuning. Building on this, we develop MATRIX Agent, a controller finetuned on M-TRACE for step-wise tool reasoning. To achieve finer alignment, we further introduce Pref-X, a set of 11K automatically generated preference pairs, and optimize MATRIX on it via step-wise preference learning. Across three benchmarks, Agent-X, GTA, and GAIA, MATRIX consistently surpasses both open- and closed-source VLMs, demonstrating scalable and effective multimodal tool use. Our data and code is avaliable at https://github.com/mbzuai-oryx/MATRIX.

Shenxu Chang, Junchi Yu, Weixing Wang et al.

Diffusion large language models (D-LLMs) have recently emerged as a promising alternative to auto-regressive LLMs (AR-LLMs). However, the hallucination problem in D-LLMs remains underexplored, limiting their reliability in real-world applications. Existing hallucination detection methods are designed for AR-LLMs and rely on signals from single-step generation, making them ill-suited for D-LLMs where hallucination signals often emerge throughout the multi-step denoising process. To bridge this gap, we propose TraceDet, a novel framework that explicitly leverages the intermediate denoising steps of D-LLMs for hallucination detection. TraceDet models the denoising process as an action trace, with each action defined as the model's prediction over the cleaned response, conditioned on the previous intermediate output. By identifying the sub-trace that is maximally informative to the hallucinated responses, TraceDet leverages the key hallucination signals in the multi-step denoising process of D-LLMs for hallucination detection. Extensive experiments on various open source D-LLMs demonstrate that TraceDet consistently improves hallucination detection, achieving an average gain in AUROC of 15.2% compared to baselines.

Haochen Luo, Jindong Gu, Fengyuan Liu et al.

Different from traditional task-specific vision models, recent large VLMs can readily adapt to different vision tasks by simply using different textual instructions, i.e., prompts. However, a well-known concern about traditional task-specific vision models is that they can be misled by imperceptible adversarial perturbations. Furthermore, the concern is exacerbated by the phenomenon that the same adversarial perturbations can fool different task-specific models. Given that VLMs rely on prompts to adapt to different tasks, an intriguing question emerges: Can a single adversarial image mislead all predictions of VLMs when a thousand different prompts are given? This question essentially introduces a novel perspective on adversarial transferability: cross-prompt adversarial transferability. In this work, we propose the Cross-Prompt Attack (CroPA). This proposed method updates the visual adversarial perturbation with learnable prompts, which are designed to counteract the misleading effects of the adversarial image. By doing this, CroPA significantly improves the transferability of adversarial examples across prompts. Extensive experiments are conducted to verify the strong cross-prompt adversarial transferability of CroPA with prevalent VLMs including Flamingo, BLIP-2, and InstructBLIP in various different tasks. Our source code is available at \url{https://github.com/Haochen-Luo/CroPA}.

Kuofeng Gao, Yang Bai, Jindong Gu et al.

Large vision-language models (VLMs) such as GPT-4 have achieved exceptional performance across various multi-modal tasks. However, the deployment of VLMs necessitates substantial energy consumption and computational resources. Once attackers maliciously induce high energy consumption and latency time (energy-latency cost) during inference of VLMs, it will exhaust computational resources. In this paper, we explore this attack surface about availability of VLMs and aim to induce high energy-latency cost during inference of VLMs. We find that high energy-latency cost during inference of VLMs can be manipulated by maximizing the length of generated sequences. To this end, we propose verbose images, with the goal of crafting an imperceptible perturbation to induce VLMs to generate long sentences during inference. Concretely, we design three loss objectives. First, a loss is proposed to delay the occurrence of end-of-sequence (EOS) token, where EOS token is a signal for VLMs to stop generating further tokens. Moreover, an uncertainty loss and a token diversity loss are proposed to increase the uncertainty over each generated token and the diversity among all tokens of the whole generated sequence, respectively, which can break output dependency at token-level and sequence-level. Furthermore, a temporal weight adjustment algorithm is proposed, which can effectively balance these losses. Extensive experiments demonstrate that our verbose images can increase the length of generated sequences by 7.87 times and 8.56 times compared to original images on MS-COCO and ImageNet datasets, which presents potential challenges for various applications. Our code is available at https://github.com/KuofengGao/Verbose_Images.

Ameya Prabhu, Zhipeng Cai, Puneet Dokania et al.

Traditional online continual learning (OCL) research has primarily focused on mitigating catastrophic forgetting with fixed and limited storage allocation throughout an agent's lifetime. However, a broad range of real-world applications are primarily constrained by computational costs rather than storage limitations. In this paper, we target such applications, investigating the online continual learning problem under relaxed storage constraints and limited computational budgets. We contribute a simple algorithm, which updates a kNN classifier continually along with a fixed, pretrained feature extractor. We selected this algorithm due to its exceptional suitability for online continual learning. It can adapt to rapidly changing streams, has zero stability gap, operates within tiny computational budgets, has low storage requirements by only storing features, and has a consistency property: It never forgets previously seen data. These attributes yield significant improvements, allowing our proposed algorithm to outperform existing methods by over 20% in accuracy on two large-scale OCL datasets: Continual LOCalization (CLOC) with 39M images and 712 classes and Continual Google Landmarks V2 (CGLM) with 580K images and 10,788 classes, even when existing methods retain all previously seen images. Furthermore, we achieve this superior performance with considerably reduced computational and storage expenses. We provide code to reproduce our results at github.com/drimpossible/ACM.

Zitong Yu, Yuming Shen, Jingang Shi et al.

Remote photoplethysmography (rPPG), which aims at measuring heart activities and physiological signals from facial video without any contact, has great potential in many applications (e.g., remote healthcare and affective computing). Recent deep learning approaches focus on mining subtle rPPG clues using convolutional neural networks with limited spatio-temporal receptive fields, which neglect the long-range spatio-temporal perception and interaction for rPPG modeling. In this paper, we propose the PhysFormer, an end-to-end video transformer based architecture, to adaptively aggregate both local and global spatio-temporal features for rPPG representation enhancement. As key modules in PhysFormer, the temporal difference transformers first enhance the quasi-periodic rPPG features with temporal difference guided global attention, and then refine the local spatio-temporal representation against interference. Furthermore, we also propose the label distribution learning and a curriculum learning inspired dynamic constraint in frequency domain, which provide elaborate supervisions for PhysFormer and alleviate overfitting. Comprehensive experiments are performed on four benchmark datasets to show our superior performance on both intra- and cross-dataset testings. One highlight is that, unlike most transformer networks needed pretraining from large-scale datasets, the proposed PhysFormer can be easily trained from scratch on rPPG datasets, which makes it promising as a novel transformer baseline for the rPPG community. The codes will be released at https://github.com/ZitongYu/PhysFormer.

Jie-Neng Chen, Shuyang Sun, Ju He et al.

Mixup-based augmentation has been found to be effective for generalizing models during training, especially for Vision Transformers (ViTs) since they can easily overfit. However, previous mixup-based methods have an underlying prior knowledge that the linearly interpolated ratio of targets should be kept the same as the ratio proposed in input interpolation. This may lead to a strange phenomenon that sometimes there is no valid object in the mixed image due to the random process in augmentation but there is still response in the label space. To bridge such gap between the input and label spaces, we propose TransMix, which mixes labels based on the attention maps of Vision Transformers. The confidence of the label will be larger if the corresponding input image is weighted higher by the attention map. TransMix is embarrassingly simple and can be implemented in just a few lines of code without introducing any extra parameters and FLOPs to ViT-based models. Experimental results show that our method can consistently improve various ViT-based models at scales on ImageNet classification. After pre-trained with TransMix on ImageNet, the ViT-based models also demonstrate better transferability to semantic segmentation, object detection and instance segmentation. TransMix also exhibits to be more robust when evaluating on 4 different benchmarks. Code will be made publicly available at https://github.com/Beckschen/TransMix.

Shiyu Tang, Ruihao Gong, Yan Wang et al.

Deep neural networks (DNNs) are vulnerable to adversarial noises, which motivates the benchmark of model robustness. Existing benchmarks mainly focus on evaluating defenses, but there are no comprehensive studies of how architecture design and training techniques affect robustness. Comprehensively benchmarking their relationships is beneficial for better understanding and developing robust DNNs. Thus, we propose RobustART, the first comprehensive Robustness investigation benchmark on ImageNet regarding ARchitecture design (49 human-designed off-the-shelf architectures and 1200+ networks from neural architecture search) and Training techniques (10+ techniques, e.g., data augmentation) towards diverse noises (adversarial, natural, and system noises). Extensive experiments substantiated several insights for the first time, e.g., (1) adversarial training is effective for the robustness against all noises types for Transformers and MLP-Mixers; (2) given comparable model sizes and aligned training settings, CNNs > Transformers > MLP-Mixers on robustness against natural and system noises; Transformers > MLP-Mixers > CNNs on adversarial robustness; (3) for some light-weight architectures, increasing model sizes or using extra data cannot improve robustness. Our benchmark presents: (1) an open-source platform for comprehensive robustness evaluation; (2) a variety of pre-trained models to facilitate robustness evaluation; and (3) a new view to better understand the mechanism towards designing robust DNNs. We will continuously develop to this ecosystem for the community.

Xiaoyu Yue, Shuyang Sun, Zhanghui Kuang et al.

Transformers with powerful global relation modeling abilities have been introduced to fundamental computer vision tasks recently. As a typical example, the Vision Transformer (ViT) directly applies a pure transformer architecture on image classification, by simply splitting images into tokens with a fixed length, and employing transformers to learn relations between these tokens. However, such naive tokenization could destruct object structures, assign grids to uninterested regions such as background, and introduce interference signals. To mitigate the above issues, in this paper, we propose an iterative and progressive sampling strategy to locate discriminative regions. At each iteration, embeddings of the current sampling step are fed into a transformer encoder layer, and a group of sampling offsets is predicted to update the sampling locations for the next step. The progressive sampling is differentiable. When combined with the Vision Transformer, the obtained PS-ViT network can adaptively learn where to look. The proposed PS-ViT is both effective and efficient. When trained from scratch on ImageNet, PS-ViT performs 3.8% higher than the vanilla ViT in terms of top-1 accuracy with about $4\times$ fewer parameters and $10\times$ fewer FLOPs. Code is available at https://github.com/yuexy/PS-ViT.

Shuyang Sun, Xiaoyu Yue, Song Bai et al.

Human vision is able to capture the part-whole hierarchical information from the entire scene. This paper presents the Visual Parser (ViP) that explicitly constructs such a hierarchy with transformers. ViP divides visual representations into two levels, the part level and the whole level. Information of each part represents a combination of several independent vectors within the whole. To model the representations of the two levels, we first encode the information from the whole into part vectors through an attention mechanism, then decode the global information within the part vectors back into the whole representation. By iteratively parsing the two levels with the proposed encoder-decoder interaction, the model can gradually refine the features on both levels. Experimental results demonstrate that ViP can achieve very competitive performance on three major tasks e.g. classification, detection and instance segmentation. In particular, it can surpass the previous state-of-the-art CNN backbones by a large margin on object detection. The tiny model of the ViP family with $7.2\times$ fewer parameters and $10.9\times$ fewer FLOPS can perform comparably with the largest model ResNeXt-101-64$\times$4d of ResNe(X)t family. Visualization results also demonstrate that the learnt parts are highly informative of the predicting class, making ViP more explainable than previous fundamental architectures. Code is available at https://github.com/kevin-ssy/ViP.

Yansong Tang, Zhenyu Jiang, Zhenda Xie et al.

Previous cycle-consistency correspondence learning methods usually leverage image patches for training. In this paper, we present a fully convolutional method, which is simpler and more coherent to the inference process. While directly applying fully convolutional training results in model collapse, we study the underline reason behind this collapse phenomenon, indicating that the absolute positions of pixels provide a shortcut to easily accomplish cycle-consistence, which hinders the learning of meaningful visual representations. To break this absolute position shortcut, we propose to apply different crops for forward and backward frames, and adopt feature warping to establish correspondence between two crops of a same frame. The former technique enforces the corresponding pixels at forward and back tracks to have different absolute positions, and the latter effectively blocks the shortcuts going between forward and back tracks. In three label propagation benchmarks for pose tracking, face landmark tracking and video object segmentation, our method largely improves the results of vanilla fully convolutional cycle-consistency method, achieving very competitive performance compared with the self-supervised state-of-the-art approaches. Our trained model and code are available at \url{https://github.com/Steve-Tod/STFC3}.