Sen He

Sen He, Yi-Zhe Song, Tao Xiang

Image-based virtual try-on aims to fit an in-shop garment into a clothed person image. To achieve this, a key step is garment warping which spatially aligns the target garment with the corresponding body parts in the person image. Prior methods typically adopt a local appearance flow estimation model. They are thus intrinsically susceptible to difficult body poses/occlusions and large mis-alignments between person and garment images (see Fig.~\ref{fig:fig1}). To overcome this limitation, a novel global appearance flow estimation model is proposed in this work. For the first time, a StyleGAN based architecture is adopted for appearance flow estimation. This enables us to take advantage of a global style vector to encode a whole-image context to cope with the aforementioned challenges. To guide the StyleGAN flow generator to pay more attention to local garment deformation, a flow refinement module is introduced to add local context. Experiment results on a popular virtual try-on benchmark show that our method achieves new state-of-the-art performance. It is particularly effective in a `in-the-wild' application scenario where the reference image is full-body resulting in a large mis-alignment with the garment image (Fig.~\ref{fig:fig1} Top). Code is available at: \url{https://github.com/SenHe/Flow-Style-VTON}.

Michał Stypułkowski, Konstantinos Vougioukas, Sen He et al.

Talking face generation has historically struggled to produce head movements and natural facial expressions without guidance from additional reference videos. Recent developments in diffusion-based generative models allow for more realistic and stable data synthesis and their performance on image and video generation has surpassed that of other generative models. In this work, we present an autoregressive diffusion model that requires only one identity image and audio sequence to generate a video of a realistic talking human head. Our solution is capable of hallucinating head movements, facial expressions, such as blinks, and preserving a given background. We evaluate our model on two different datasets, achieving state-of-the-art results on both of them.

Xiao Han, Sen He, Li Zhang et al.

Interactive garment retrieval (IGR) aims to retrieve a target garment image based on a reference garment image along with user feedback on what to change on the reference garment. Two IGR tasks have been studied extensively: text-guided garment retrieval (TGR) and visually compatible garment retrieval (VCR). The user feedback for the former indicates what semantic attributes to change with the garment category preserved, while the category is the only thing to be changed explicitly for the latter, with an implicit requirement on style preservation. Despite the similarity between these two tasks and the practical need for an efficient system tackling both, they have never been unified and modeled jointly. In this paper, we propose a Unified Interactive Garment Retrieval (UIGR) framework to unify TGR and VCR. To this end, we first contribute a large-scale benchmark suited for both problems. We further propose a strong baseline architecture to integrate TGR and VCR in one model. Extensive experiments suggest that unifying two tasks in one framework is not only more efficient by requiring a single model only, it also leads to better performance. Code and datasets are available at https://github.com/BrandonHanx/CompFashion.

Zhihe Lu, Sen He, Da Li et al.

Generalized Few-shot Semantic Segmentation (GFSS) aims to segment each image pixel into either base classes with abundant training examples or novel classes with only a handful of (e.g., 1-5) training images per class. Compared to the widely studied Few-shot Semantic Segmentation FSS, which is limited to segmenting novel classes only, GFSS is much under-studied despite being more practical. Existing approach to GFSS is based on classifier parameter fusion whereby a newly trained novel class classifier and a pre-trained base class classifier are combined to form a new classifier. As the training data is dominated by base classes, this approach is inevitably biased towards the base classes. In this work, we propose a novel Prediction Calibration Network PCN to address this problem. Instead of fusing the classifier parameters, we fuse the scores produced separately by the base and novel classifiers. To ensure that the fused scores are not biased to either the base or novel classes, a new Transformer-based calibration module is introduced. It is known that the lower-level features are useful of detecting edge information in an input image than higher-level features. Thus, we build a cross-attention module that guides the classifier's final prediction using the fused multi-level features. However, transformers are computationally demanding. Crucially, to make the proposed cross-attention module training tractable at the pixel level, this module is designed based on feature-score cross-covariance and episodically trained to be generalizable at inference time. Extensive experiments on PASCAL-$5^{i}$ and COCO-$20^{i}$ show that our PCN outperforms the state-the-the-art alternatives by large margins.

Yuren Cong, Mengmeng Xu, Christian Simon et al.

Text-to-video editing aims to edit the visual appearance of a source video conditional on textual prompts. A major challenge in this task is to ensure that all frames in the edited video are visually consistent. Most recent works apply advanced text-to-image diffusion models to this task by inflating 2D spatial attention in the U-Net into spatio-temporal attention. Although temporal context can be added through spatio-temporal attention, it may introduce some irrelevant information for each patch and therefore cause inconsistency in the edited video. In this paper, for the first time, we introduce optical flow into the attention module in the diffusion model's U-Net to address the inconsistency issue for text-to-video editing. Our method, FLATTEN, enforces the patches on the same flow path across different frames to attend to each other in the attention module, thus improving the visual consistency in the edited videos. Additionally, our method is training-free and can be seamlessly integrated into any diffusion-based text-to-video editing methods and improve their visual consistency. Experiment results on existing text-to-video editing benchmarks show that our proposed method achieves the new state-of-the-art performance. In particular, our method excels in maintaining the visual consistency in the edited videos.

Sen He, Yi-Zhe Song, Tao Xiang

Multi-pose virtual try-on (MPVTON) aims to fit a target garment onto a person at a target pose. Compared to traditional virtual try-on (VTON) that fits the garment but keeps the pose unchanged, MPVTON provides a better try-on experience, but is also more challenging due to the dual garment and pose editing objectives. Existing MPVTON methods adopt a pipeline comprising three disjoint modules including a target semantic layout prediction module, a coarse try-on image generator and a refinement try-on image generator. These models are trained separately, leading to sub-optimal model training and unsatisfactory results. In this paper, we propose a novel single stage model for MPVTON. Key to our model is a parallel flow estimation module that predicts the flow fields for both person and garment images conditioned on the target pose. The predicted flows are subsequently used to warp the appearance feature maps of the person and the garment images to construct a style map. The map is then used to modulate the target pose's feature map for target try-on image generation. With the parallel flow estimation design, our model can be trained end-to-end in a single stage and is more computationally efficient, resulting in new SOTA performance on existing MPVTON benchmarks. We further introduce multi-task training and demonstrate that our model can also be applied for traditional VTON and pose transfer tasks and achieve comparable performance to SOTA specialized models on both tasks.

Aiyu Cui, Sen He, Tao Xiang et al.

Virtual try-on, i.e making people virtually try new garments, is an active research area in computer vision with great commercial applications. Current virtual try-on methods usually work in a two-stage pipeline. First, the garment image is warped on the person's pose using a flow estimation network. Then in the second stage, the warped garment is fused with the person image to render a new try-on image. Unfortunately, such methods are heavily dependent on the quality of the garment warping which often fails when dealing with hard poses (e.g., a person lifting or crossing arms). In this work, we propose a robust warping method for virtual try-on based on a learned garment DensePose which has a direct correspondence with the person's DensePose. Due to the lack of annotated data, we show how to leverage an off-the-shelf person DensePose model and a pretrained flow model to learn the garment DensePose in a weakly supervised manner. The garment DensePose allows a robust warping to any person's pose without any additional computation. Our method achieves the state-of-the-art equivalent on virtual try-on benchmarks and shows warping robustness on in-the-wild person images with hard poses, making it more suited for real-world virtual try-on applications.

Zhaochong An, Menglin Jia, Haonan Qiu et al.

Storytelling in real-world videos often unfolds through multiple shots -- discontinuous yet semantically connected clips that together convey a coherent narrative. However, existing multi-shot video generation (MSV) methods struggle to effectively model long-range cross-shot context, as they rely on limited temporal windows or single keyframe conditioning, leading to degraded performance under complex narratives. In this work, we propose OneStory, enabling global yet compact cross-shot context modeling for consistent and scalable narrative generation. OneStory reformulates MSV as a next-shot generation task, enabling autoregressive shot synthesis while leveraging pretrained image-to-video (I2V) models for strong visual conditioning. We introduce two key modules: a Frame Selection module that constructs a semantically-relevant global memory based on informative frames from prior shots, and an Adaptive Conditioner that performs importance-guided patchification to generate compact context for direct conditioning. We further curate a high-quality multi-shot dataset with referential captions to mirror real-world storytelling patterns, and design effective training strategies under the next-shot paradigm. Finetuned from a pretrained I2V model on our curated 60K dataset, OneStory achieves state-of-the-art narrative coherence across diverse and complex scenes in both text- and image-conditioned settings, enabling controllable and immersive long-form video storytelling.

Zhiheng Liu, Weiming Ren, Haozhe Liu et al.

Unified multimodal models (UMMs) aim to jointly perform multimodal understanding and generation within a single framework. We present TUNA, a native UMM that builds a unified continuous visual representation by cascading a VAE encoder with a representation encoder. This unified representation space allows end-to-end processing of images and videos for both understanding and generation tasks. Compared to prior UMMs with decoupled representations, TUNA's unified visual space avoids representation format mismatches introduced by separate encoders, outperforming decoupled alternatives in both understanding and generation. Moreover, we observe that stronger pretrained representation encoders consistently yield better performance across all multimodal tasks, highlighting the importance of the representation encoder. Finally, in this unified setting, jointly training on both understanding and generation data allows the two tasks to benefit from each other rather than interfere. Our extensive experiments on multimodal understanding and generation benchmarks show that TUNA achieves state-of-the-art results in image and video understanding, image and video generation, and image editing, demonstrating the effectiveness and scalability of its unified representation design.

Haonan Qiu, Shikun Liu, Zijian Zhou et al.

High-resolution video generation, while crucial for digital media and film, is computationally bottlenecked by the quadratic complexity of diffusion models, making practical inference infeasible. To address this, we introduce HiStream, an efficient autoregressive framework that systematically reduces redundancy across three axes: i) Spatial Compression: denoising at low resolution before refining at high resolution with cached features; ii) Temporal Compression: a chunk-by-chunk strategy with a fixed-size anchor cache, ensuring stable inference speed; and iii) Timestep Compression: applying fewer denoising steps to subsequent, cache-conditioned chunks. On 1080p benchmarks, our primary HiStream model (i+ii) achieves state-of-the-art visual quality while demonstrating up to 76.2x faster denoising compared to the Wan2.1 baseline and negligible quality loss. Our faster variant, HiStream+, applies all three optimizations (i+ii+iii), achieving a 107.5x acceleration over the baseline, offering a compelling trade-off between speed and quality, thereby making high-resolution video generation both practical and scalable.

Zhenyu Qi, Qian Zhang, Haotang Li et al.

Quantum compilers rely on calibration-derived noise models to guide circuit mapping and optimization. These models characterize gate and qubit errors independently and miss context-dependent effects such as crosstalk and correlated scheduling errors. As a result, two compiled circuits that score equally under the noise model can behave very differently on real hardware, and the compiler has no mechanism to learn from such recurring mismatches. We present QRisk, a framework that discovers backend-specific abnormal patterns from real hardware executions. QRisk uses delta debugging to isolate compact circuit fragments that consistently produce excess error not predicted by the noise model, then validates their persistence across repeated runs and calibration windows. The verified patterns are stored in a backend-specific pattern database. At compilation time, QRisk scans a compiled circuit for occurrences of known patterns and applies targeted commuting gate swaps to disrupt them, producing a semantically equivalent circuit with fewer abnormal patterns. We evaluate QRisk on two IBM backends (ibm_fez and ibm_marrakesh) using Grover search circuits. On both backends, discovered patterns persist across multiple calibration windows over months. Disrupting these patterns via commuting gate swaps reduces excess hardware noise by 24% on ibm_fez (Spearman $ρ$ = 0.515, p = 0.0007) and 45% on ibm_marrakesh ($ρ$ = 0.711, p < 0.0001), while the noise model predicts identical error for all equivalent circuits. Testing on a third backend confirms that these patterns are backend-specific.

Haozhe Liu, Ding Liu, Mingchen Zhuge et al.

We introduce MoS (Mixture of States), a novel fusion paradigm for multimodal diffusion models that merges modalities using flexible, state-based interactions. The core of MoS is a learnable, token-wise router that creates denoising timestep- and input-dependent interactions between modalities' hidden states, precisely aligning token-level features with the diffusion trajectory. This router sparsely selects the top-$k$ hidden states and is trained with an $ε$-greedy strategy, efficiently selecting contextual features with minimal learnable parameters and negligible computational overhead. We validate our design with text-to-image generation (MoS-Image) and editing (MoS-Editing), which achieve state-of-the-art results. With only 3B to 5B parameters, our models match or surpass counterparts up to $4\times$ larger. These findings establish MoS as a flexible and compute-efficient paradigm for scaling multimodal diffusion models.

Zijian Zhou, Shikun Liu, Haozhe Liu et al.

Reference-to-video (R2V) generation aims to synthesize videos that align with a text prompt while preserving the subject identity from reference images. However, current R2V methods are hindered by the reliance on explicit reference image-video-text triplets, whose construction is highly expensive and difficult to scale. We bypass this bottleneck by introducing Saber, a scalable zero-shot framework that requires no explicit R2V data. Trained exclusively on video-text pairs, Saber employs a masked training strategy and a tailored attention-based model design to learn identity-consistent and reference-aware representations. Mask augmentation techniques are further integrated to mitigate copy-paste artifacts common in reference-to-video generation. Moreover, Saber demonstrates remarkable generalization capabilities across a varying number of references and achieves superior performance on the OpenS2V-Eval benchmark compared to methods trained with R2V data.

Huashan Chen, Zhenyu Qi, Haotang Li et al.

Large Language Models (LLMs) have become central to automated code generation, yet existing approaches operate within a single-LLM paradigm: one model is selected and applied throughout the entire generation process. We observe that different LLMs exhibit complementary strengths: no single model dominates across all programming languages, algorithmic problem categories, or development stages. Multi-LLM collaboration, structured as per-stage, per-category routing rather than majority voting, produces higher-quality code than any individual model. Based on this observation, we propose PerfOrch, a multi-agent orchestration system that decomposes code generation into four collaborative agents: categorization, generation, debugging, and refinement. Each agent maintains a Memory module: a ranking matrix indexed by programming language and problem category, constructed from offline profiling and consulted at runtime to select the most suitable model for each task. We evaluate PerfOrch on two benchmarks, HumanEval-X and EffiBench-X, totaling 2,500 problems across five languages (Python, Java, C++, Go, and Rust). PerfOrch achieves average pass@1 rates of 97.19% on HumanEval-X and 95.83% on EffiBench-X, improving over the strongest single-model pipeline by 1.22-14.58 percentage points across languages. Notably, Memory rankings constructed solely from HumanEval-X profiling generalize to the entirely unseen EffiBench-X benchmark without re-profiling, demonstrating that the complementary-strength patterns PerfOrch exploits are properties of the models rather than artifacts of a specific problem distribution. Beyond correctness, PerfOrch improves execution time for 61-90% of solved problems with mean speedups of 4.7-29.9%, matching the refinement coverage of exhaustive multi-model evaluation at roughly half the token cost.

Xiaoke Huang, Bhavul Gauri, Kam Woh Ng et al.

Vector glyphs are the atomic units of digital typography, yet most learning-based pipelines still depend on carefully curated exemplar sheets and raster-to-vector postprocessing, which limits accessibility and editability. We introduce VecGlypher, a single multimodal language model that generates high-fidelity vector glyphs directly from text descriptions or image exemplars. Given a style prompt, optional reference glyph images, and a target character, VecGlypher autoregressively emits SVG path tokens, avoiding raster intermediates and producing editable, watertight outlines in one pass. A typography-aware data and training recipe makes this possible: (i) a large-scale continuation stage on 39K noisy Envato fonts to master SVG syntax and long-horizon geometry, followed by (ii) post-training on 2.5K expert-annotated Google Fonts with descriptive tags and exemplars to align language and imagery with geometry; preprocessing normalizes coordinate frames, canonicalizes paths, de-duplicates families, and quantizes coordinates for stable long-sequence decoding. On cross-family OOD evaluation, VecGlypher substantially outperforms both general-purpose LLMs and specialized vector-font baselines for text-only generation, while image-referenced generation reaches a state-of-the-art performance, with marked gains over DeepVecFont-v2 and DualVector. Ablations show that model scale and the two-stage recipe are critical and that absolute-coordinate serialization yields the best geometry. VecGlypher lowers the barrier to font creation by letting users design with words or exemplars, and provides a scalable foundation for future multimodal design tools.

Fei Zhang, Zijian Zhou, Bohao Tang et al.

We introduce the first method, to the best of our knowledge, for adapting image-to-video models to layer-aware text (glyph) animation, a capability critical for practical dynamic visual design. Existing approaches predominantly handle the transparency-encoding (alpha channel) as an extra latent dimension appended to the RGB space, necessitating the reconstruction of the underlying RGB-centric variational autoencoder (VAE). However, given the scarcity of high-quality transparent glyph data, retraining the VAE is computationally expensive and may erode the robust semantic priors learned from massive RGB corpora, potentially leading to latent pattern mixing. To mitigate these limitations, we propose TransText, a framework based on a novel Alpha-as-RGB paradigm to jointly model appearance and transparency without modifying the pre-trained generative manifold. TransText embeds the alpha channel as an RGB-compatible visual signal through latent spatial concatenation, explicitly ensuring strict cross-modal (RGB-and-Alpha) consistency while preventing feature entanglement. Our experiments demonstrate that TransText significantly outperforms baselines, generating coherent, high-fidelity transparent animations with diverse, fine-grained effects.

Shuo Ju, Qingzhao Zhang, Huashan Chen et al.

Existing physical adversarial attacks on vision-based autonomous driving induce time-evolving perception errors, including biased object tracking or trajectory prediction, through (i) sophisticated physical patch inducing detection box drift when entering the view distance, or (ii) dynamically changing patches that cause different perception errors at different time. In both cases, viewing-angle variation is treated as a challenge, requiring adversarial patches to remain effective across frames under varying views, leading to complex multi-view optimization. In contrast, we show that viewing-angle variation itself can be turned into an attack tool. We design a new attack paradigm where a static, passive adversarial camouflage is mounted on a vehicle whose view-dependent appearance naturally evolves with relative motion, inducing consistent feature drift across frames. This causes the system to infer a physically plausible but incorrect trajectory, such as a false cut-in, which propagates to downstream decision-making and triggers unnecessary braking. Unlike prior approaches that require multi-view robustness or active intervention, our attack emerges from normal driving dynamics and is easy to deploy: a parked vehicle with a natural camouflage can induce hard braking in passing autonomous vehicles. We demonstrate the novel attack on nuScenes dataset, showing the effectiveness with an end-to-end success rate of up to 87.5%, measured by hard-braking events, and robustness across different scene backgrounds, victim vehicle speeds, and perception models.

Xiao Han, Sen He, Li Zhang et al.

Text-based person search (TBPS) aims at retrieving a target person from an image gallery with a descriptive text query. Solving such a fine-grained cross-modal retrieval task is challenging, which is further hampered by the lack of large-scale datasets. In this paper, we present a framework with two novel components to handle the problems brought by limited data. Firstly, to fully utilize the existing small-scale benchmarking datasets for more discriminative feature learning, we introduce a cross-modal momentum contrastive learning framework to enrich the training data for a given mini-batch. Secondly, we propose to transfer knowledge learned from existing coarse-grained large-scale datasets containing image-text pairs from drastically different problem domains to compensate for the lack of TBPS training data. A transfer learning method is designed so that useful information can be transferred despite the large domain gap. Armed with these components, our method achieves new state of the art on the CUHK-PEDES dataset with significant improvements over the prior art in terms of Rank-1 and mAP. Our code is available at https://github.com/BrandonHanx/TextReID.

Zhihe Lu, Sen He, Xiatian Zhu et al.

A few-shot semantic segmentation model is typically composed of a CNN encoder, a CNN decoder and a simple classifier (separating foreground and background pixels). Most existing methods meta-learn all three model components for fast adaptation to a new class. However, given that as few as a single support set image is available, effective model adaption of all three components to the new class is extremely challenging. In this work we propose to simplify the meta-learning task by focusing solely on the simplest component, the classifier, whilst leaving the encoder and decoder to pre-training. We hypothesize that if we pre-train an off-the-shelf segmentation model over a set of diverse training classes with sufficient annotations, the encoder and decoder can capture rich discriminative features applicable for any unseen classes, rendering the subsequent meta-learning stage unnecessary. For the classifier meta-learning, we introduce a Classifier Weight Transformer (CWT) designed to dynamically adapt the supportset trained classifier's weights to each query image in an inductive way. Extensive experiments on two standard benchmarks show that despite its simplicity, our method outperforms the state-of-the-art alternatives, often by a large margin.Code is available on https://github.com/zhiheLu/CWT-for-FSS.

Sen He, Hamed R. Tavakoli, Ali Borji et al.

In this work, we present a novel dataset consisting of eye movements and verbal descriptions recorded synchronously over images. Using this data, we study the differences in human attention during free-viewing and image captioning tasks. We look into the relationship between human attention and language constructs during perception and sentence articulation. We also analyse attention deployment mechanisms in the top-down soft attention approach that is argued to mimic human attention in captioning tasks, and investigate whether visual saliency can help image captioning. Our study reveals that (1) human attention behaviour differs in free-viewing and image description tasks. Humans tend to fixate on a greater variety of regions under the latter task, (2) there is a strong relationship between described objects and attended objects ($97\%$ of the described objects are being attended), (3) a convolutional neural network as feature encoder accounts for human-attended regions during image captioning to a great extent (around $78\%$), (4) soft-attention mechanism differs from human attention, both spatially and temporally, and there is low correlation between caption scores and attention consistency scores. These indicate a large gap between humans and machines in regards to top-down attention, and (5) by integrating the soft attention model with image saliency, we can significantly improve the model's performance on Flickr30k and MSCOCO benchmarks. The dataset can be found at: https://github.com/SenHe/Human-Attention-in-Image-Captioning.

Shoufa Chen, Mengmeng Xu, Jiawei Ren et al.

In this study, we explore Transformer-based diffusion models for image and video generation. Despite the dominance of Transformer architectures in various fields due to their flexibility and scalability, the visual generative domain primarily utilizes CNN-based U-Net architectures, particularly in diffusion-based models. We introduce GenTron, a family of Generative models employing Transformer-based diffusion, to address this gap. Our initial step was to adapt Diffusion Transformers (DiTs) from class to text conditioning, a process involving thorough empirical exploration of the conditioning mechanism. We then scale GenTron from approximately 900M to over 3B parameters, observing significant improvements in visual quality. Furthermore, we extend GenTron to text-to-video generation, incorporating novel motion-free guidance to enhance video quality. In human evaluations against SDXL, GenTron achieves a 51.1% win rate in visual quality (with a 19.8% draw rate), and a 42.3% win rate in text alignment (with a 42.9% draw rate). GenTron also excels in the T2I-CompBench, underscoring its strengths in compositional generation. We believe this work will provide meaningful insights and serve as a valuable reference for future research.

Haotang Li, Yili Ren, Zhenyu Qi et al.

Body fat percentage and its spatial distribution are clinically important health indicators. However, existing measurement methods often impose a tradeoff between accuracy and accessibility. Clinical-grade techniques, such as Dual-Energy X-ray Absorptiometry (DEXA) and hydrostatic weighing, provide accurate measurements but require specialized equipment and trained operators, making them difficult to access and unsuitable for everyday use. In contrast, consumer-level methods, such as Bioelectrical Impedance Analysis (BIA) smart scales and skinfold calipers, are more accessible but typically provide only coarse-grained estimates, are prone to user error, or require intrusive physical contact. In this work, we present UWB-Fat, the first system that leverages commodity ultra-wideband (UWB) radar to enable non-intrusive, accessible, and accurate caliper-equivalent skinfold thickness estimation, serving as a convenient replacement for the skinfold caliper. UWB-Fat collects UWB signal at specified body sites non-intrusively without operator assistance. It extracts body-composition-related features from UWB signals by exploiting dielectric contrasts among skin, fat, and muscle tissues. Then, it uses a physics-inspired model to estimate site-specific skinfold thickness. We evaluate UWB-Fat on 15 participants, achieving a root mean square error of 0.63~mm for pooled-site subcutaneous fat thickness. These results highlight the potential of UWB-Fat to support low-cost, self-administered, and everyday body fat monitoring.

Haotang Li, Zhenyu Qi, Shaohan Henry Wang et al.

Visual Geometry Transformer (VGGT) is a strong feed-forward model for multiple 3D tasks, but its Alternating-Attention (AA) stack scales quadratically in the total token count, making long clips expensive. Existing token-reduction accelerators operate inside AA, leaving the patch grid that enters AA uncompressed. We introduce PaceVGGT, a pre-AA token pruning framework that prunes DINO patch tokens before the first AA block of a frozen VGGT. PaceVGGT trains a lightweight Token Scorer that estimates per-token importance from DINO features. The scorer is first distilled against an AA-internal attention target from the unpruned backbone, then refined under downstream camera, depth, and point-map losses. A per-frame keep budget fixes the backbone-visible sequence length, while an importance-adaptive merge/prune assignment preserves residual content from high-saliency frames under a fixed total merge budget. A Feature-guided Restoration module reconstructs the dense spatial grid required by the prediction heads. On ScanNet-50 and 7-Scenes, PaceVGGT remains on the reconstruction quality--latency frontier while reducing inference latency. On ScanNet-50, it reduces latency by \(5.1\times\) over unmodified VGGT at \(N=300\) and \(1.47\times\) over LiteVGGT at \(N=1000\). These results identify pre-AA pruning as a viable acceleration route for frozen VGGT-style geometry transformers.

Zhiheng Liu, Weiming Ren, Xiaoke Huang et al.

Unified multimodal models typically rely on pretrained vision encoders and use separate visual representations for understanding and generation, creating misalignment between the two tasks and preventing fully end-to-end optimization from raw pixels. We introduce Tuna-2, a native unified multimodal model that performs visual understanding and generation directly based on pixel embeddings. Tuna-2 drastically simplifies the model architecture by employing simple patch embedding layers to encode visual input, completely discarding the modular vision encoder designs such as the VAE or the representation encoder. Experiments show that Tuna-2 achieves state-of-the-art performance in multimodal benchmarks, demonstrating that unified pixel-space modelling can fully compete with latent-space approaches for high-quality image generation. Moreover, while the encoder-based variant converges faster in early pretraining, Tuna-2's encoder-free design achieves stronger multimodal understanding at scale, particularly on tasks requiring fine-grained visual perception. These results show that pretrained vision encoders are not necessary for multimodal modelling, and end-to-end pixel-space learning offers a scalable path toward stronger visual representations for both generation and perception.

Kumara Kahatapitiya, Haozhe Liu, Sen He et al.

Generating temporally-consistent high-fidelity videos can be computationally expensive, especially over longer temporal spans. More-recent Diffusion Transformers (DiTs) -- despite making significant headway in this context -- have only heightened such challenges as they rely on larger models and heavier attention mechanisms, resulting in slower inference speeds. In this paper, we introduce a training-free method to accelerate video DiTs, termed Adaptive Caching (AdaCache), which is motivated by the fact that "not all videos are created equal": meaning, some videos require fewer denoising steps to attain a reasonable quality than others. Building on this, we not only cache computations through the diffusion process, but also devise a caching schedule tailored to each video generation, maximizing the quality-latency trade-off. We further introduce a Motion Regularization (MoReg) scheme to utilize video information within AdaCache, essentially controlling the compute allocation based on motion content. Altogether, our plug-and-play contributions grant significant inference speedups (e.g. up to 4.7x on Open-Sora 720p - 2s video generation) without sacrificing the generation quality, across multiple video DiT baselines.

Haozhe Liu, Shikun Liu, Zijian Zhou et al.

We introduce MarDini, a new family of video diffusion models that integrate the advantages of masked auto-regression (MAR) into a unified diffusion model (DM) framework. Here, MAR handles temporal planning, while DM focuses on spatial generation in an asymmetric network design: i) a MAR-based planning model containing most of the parameters generates planning signals for each masked frame using low-resolution input; ii) a lightweight generation model uses these signals to produce high-resolution frames via diffusion de-noising. MarDini's MAR enables video generation conditioned on any number of masked frames at any frame positions: a single model can handle video interpolation (e.g., masking middle frames), image-to-video generation (e.g., masking from the second frame onward), and video expansion (e.g., masking half the frames). The efficient design allocates most of the computational resources to the low-resolution planning model, making computationally expensive but important spatio-temporal attention feasible at scale. MarDini sets a new state-of-the-art for video interpolation; meanwhile, within few inference steps, it efficiently generates videos on par with those of much more expensive advanced image-to-video models.

Haotang Li, Zhenyu Qi, Hao Qin et al.

Self-supervised semantic segmentation methods often fail when faced with appearance ambiguities. We argue that this is due to an over-reliance on unstable, appearance-based features such as shadows, glare, and local textures. We propose \textbf{GASeg}, a novel framework that bridges appearance and geometry by leveraging stable topological information. The core of our method is Differentiable Box-Counting (\textbf{DBC}) module, which quantifies multi-scale topological statistics from two parallel streams: geometric-based features and appearance-based features. To force the model to learn these stable structural representations, we introduce Topological Augmentation (\textbf{TopoAug}), an adversarial strategy that simulates real-world ambiguities by applying morphological operators to the input images. A multi-objective loss, \textbf{GALoss}, then explicitly enforces cross-modal alignment between geometric-based and appearance-based features. Extensive experiments demonstrate that GASeg achieves state-of-the-art performance on four benchmarks, including COCO-Stuff, Cityscapes, and PASCAL, validating our approach of bridging geometry and appearance via topological information.

Zijian Zhou, Shikun Liu, Xiao Han et al.

Controllable person image generation aims to generate a person image conditioned on reference images, allowing precise control over the person's appearance or pose. However, prior methods often distort fine-grained textural details from the reference image, despite achieving high overall image quality. We attribute these distortions to inadequate attention to corresponding regions in the reference image. To address this, we thereby propose learning flow fields in attention (Leffa), which explicitly guides the target query to attend to the correct reference key in the attention layer during training. Specifically, it is realized via a regularization loss on top of the attention map within a diffusion-based baseline. Our extensive experiments show that Leffa achieves state-of-the-art performance in controlling appearance (virtual try-on) and pose (pose transfer), significantly reducing fine-grained detail distortion while maintaining high image quality. Additionally, we show that our loss is model-agnostic and can be used to improve the performance of other diffusion models.

Christian Simon, Sen He, Juan-Manuel Perez-Rua et al.

Solving image-to-3D from a single view is an ill-posed problem, and current neural reconstruction methods addressing it through diffusion models still rely on scene-specific optimization, constraining their generalization capability. To overcome the limitations of existing approaches regarding generalization and consistency, we introduce a novel neural rendering technique. Our approach employs the signed distance function as the surface representation and incorporates generalizable priors through geometry-encoding volumes and HyperNetworks. Specifically, our method builds neural encoding volumes from generated multi-view inputs. We adjust the weights of the SDF network conditioned on an input image at test-time to allow model adaptation to novel scenes in a feed-forward manner via HyperNetworks. To mitigate artifacts derived from the synthesized views, we propose the use of a volume transformer module to improve the aggregation of image features instead of processing each viewpoint separately. Through our proposed method, dubbed as Hyper-VolTran, we avoid the bottleneck of scene-specific optimization and maintain consistency across the images generated from multiple viewpoints. Our experiments show the advantages of our proposed approach with consistent results and rapid generation.

Yicheng Leng, Syed Muhammad Anwar, Islem Rekik et al.

Functional Magnetic Resonance Imaging (fMRI) provides useful insights into the brain function both during task or rest. Representing fMRI data using correlation matrices is found to be a reliable method of analyzing the inherent connectivity of the brain in the resting and active states. Graph Neural Networks (GNNs) have been widely used for brain network analysis due to their inherent explainability capability. In this work, we introduce a novel framework using contrastive self-supervised learning graph transformers, incorporating a brain network transformer encoder with random graph alterations. The proposed network leverages both contrastive learning and graph alterations to effectively train the graph transformer for autism detection. Our approach, tested on Autism Brain Imaging Data Exchange (ABIDE) data, demonstrates superior autism detection, achieving an AUROC of 82.6 and an accuracy of 74%, surpassing current state-of-the-art methods.

Huashan Chen, Yifan Xu, Yue Feng et al.

Biometrics authentication has become increasingly popular due to its security and convenience; however, traditional biometrics are becoming less desirable in scenarios such as new mobile devices, Virtual Reality, and Smart Vehicles. For example, while face authentication is widely used, it suffers from significant privacy concerns. The collection of complete facial data makes it less desirable for privacy-sensitive applications. Lip authentication, on the other hand, has emerged as a promising biometrics method. However, existing lip-based authentication methods heavily depend on static lip shape when the mouth is closed, which can be less robust due to lip shape dynamic motion and can barely work when the user is speaking. In this paper, we revisit the nature of lip biometrics and extract shape-independent features from the lips. We study the dynamic characteristics of lip biometrics based on articulator motion. Building on the knowledge, we propose a system for shape-independent continuous authentication via lip articulator dynamics. This system enables robust, shape-independent and continuous authentication, making it particularly suitable for scenarios with high security and privacy requirements. We conducted comprehensive experiments in different environments and attack scenarios and collected a dataset of 50 subjects. The results indicate that our system achieves an overall accuracy of 99.06% and demonstrates robustness under advanced mimic attacks and AI deepfake attacks, making it a viable solution for continuous biometric authentication in various applications.

Haotang Li, Zhenyu Qi, Sen He et al.

Improper sitting posture during prolonged computer use has become a significant public health concern. Traditional posture monitoring solutions face substantial barriers, including privacy concerns with camera-based systems and user discomfort with wearable sensors. This paper presents UWB-PostureGuard, a privacy-preserving ultra-wideband (UWB) sensing system that advances mobile technologies for preventive health management through continuous, contactless monitoring of ergonomic sitting posture. Our system leverages commercial UWB devices, utilizing comprehensive feature engineering to extract multiple ergonomic sitting posture features. We develop PoseGBDT to effectively capture temporal dependencies in posture patterns, addressing limitations of traditional frame-wise classification approaches. Extensive real-world evaluation across 10 participants and 19 distinct postures demonstrates exceptional performance, achieving 99.11% accuracy while maintaining robustness against environmental variables such as clothing thickness, additional devices, and furniture configurations. Our system provides a scalable, privacy-preserving mobile health solution on existing platforms for proactive ergonomic management, improving quality of life at low costs.

Vasu Agrawal, Akinniyi Akinyemi, Kathryn Alvero et al.

Human communication involves a complex interplay of verbal and nonverbal signals, essential for conveying meaning and achieving interpersonal goals. To develop socially intelligent AI technologies, it is crucial to develop models that can both comprehend and generate dyadic behavioral dynamics. To this end, we introduce the Seamless Interaction Dataset, a large-scale collection of over 4,000 hours of face-to-face interaction footage from over 4,000 participants in diverse contexts. This dataset enables the development of AI technologies that understand dyadic embodied dynamics, unlocking breakthroughs in virtual agents, telepresence experiences, and multimodal content analysis tools. We also develop a suite of models that utilize the dataset to generate dyadic motion gestures and facial expressions aligned with human speech. These models can take as input both the speech and visual behavior of their interlocutors. We present a variant with speech from an LLM model and integrations with 2D and 3D rendering methods, bringing us closer to interactive virtual agents. Additionally, we describe controllable variants of our motion models that can adapt emotional responses and expressivity levels, as well as generating more semantically-relevant gestures. Finally, we discuss methods for assessing the quality of these dyadic motion models, which are demonstrating the potential for more intuitive and responsive human-AI interactions.

Kebin Peng, Haotang Li, Zhenyu Qi et al.

Existing self-supervised monocular depth estimation (MDE) models attempt to improve nighttime performance by using GANs to transfer nighttime images into their daytime versions. However, this can introduce inconsistencies due to the complexities of real-world daytime lighting variations, which may finally lead to inaccurate estimation results. To address this issue, we leverage physical-prior-knowledge about light wavelength and light attenuation during nighttime. Specifically, our model, Light-Attenuation-Aware Network (LAA-Net), incorporates physical insights from Rayleigh scattering theory for robust nighttime depth estimation: LAA-Net is trained based on red channel values because red light preserves more information under nighttime scenarios due to its longer wavelength. Additionally, based on Beer-Lambert law, we introduce Red Channel Attenuation (RCA) loss to guide LAA-Net's training. Experiments on the RobotCar-Night, nuScenes-Night, RobotCar-Day, and KITTI datasets demonstrate that our model outperforms SOTA models.

Tianyuan Yu, Sen He, Yi-Zhe Song et al.

Graph neural networks (GNNs) have been used to tackle the few-shot learning (FSL) problem and shown great potentials under the transductive setting. However under the inductive setting, existing GNN based methods are less competitive. This is because they use an instance GNN as a label propagation/classification module, which is jointly meta-learned with a feature embedding network. This design is problematic because the classifier needs to adapt quickly to new tasks while the embedding does not. To overcome this problem, in this paper we propose a novel hybrid GNN (HGNN) model consisting of two GNNs, an instance GNN and a prototype GNN. Instead of label propagation, they act as feature embedding adaptation modules for quick adaptation of the meta-learned feature embedding to new tasks. Importantly they are designed to deal with a fundamental yet often neglected challenge in FSL, that is, with only a handful of shots per class, any few-shot classifier would be sensitive to badly sampled shots which are either outliers or can cause inter-class distribution overlapping. %Our two GNNs are designed to address these two types of poorly sampled few-shots respectively and their complementarity is exploited in the hybrid GNN model. Extensive experiments show that our HGNN obtains new state-of-the-art on three FSL benchmarks.

Sen He, Wentong Liao, Michael Ying Yang et al.

A lifespan face synthesis (LFS) model aims to generate a set of photo-realistic face images of a person's whole life, given only one snapshot as reference. The generated face image given a target age code is expected to be age-sensitive reflected by bio-plausible transformations of shape and texture, while being identity preserving. This is extremely challenging because the shape and texture characteristics of a face undergo separate and highly nonlinear transformations w.r.t. age. Most recent LFS models are based on generative adversarial networks (GANs) whereby age code conditional transformations are applied to a latent face representation. They benefit greatly from the recent advancements of GANs. However, without explicitly disentangling their latent representations into the texture, shape and identity factors, they are fundamentally limited in modeling the nonlinear age-related transformation on texture and shape whilst preserving identity. In this work, a novel LFS model is proposed to disentangle the key face characteristics including shape, texture and identity so that the unique shape and texture age transformations can be modeled effectively. This is achieved by extracting shape, texture and identity features separately from an encoder. Critically, two transformation modules, one conditional convolution based and the other channel attention based, are designed for modeling the nonlinear shape and texture feature transformations respectively. This is to accommodate their rather distinct aging processes and ensure that our synthesized images are both age-sensitive and identity preserving. Extensive experiments show that our LFS model is clearly superior to the state-of-the-art alternatives. Codes and demo are available on our project website: \url{https://senhe.github.io/projects/iccv_2021_lifespan_face}.

Sen He, Wentong Liao, Michael Ying Yang et al.

A layout to image (L2I) generation model aims to generate a complicated image containing multiple objects (things) against natural background (stuff), conditioned on a given layout. Built upon the recent advances in generative adversarial networks (GANs), existing L2I models have made great progress. However, a close inspection of their generated images reveals two major limitations: (1) the object-to-object as well as object-to-stuff relations are often broken and (2) each object's appearance is typically distorted lacking the key defining characteristics associated with the object class. We argue that these are caused by the lack of context-aware object and stuff feature encoding in their generators, and location-sensitive appearance representation in their discriminators. To address these limitations, two new modules are proposed in this work. First, a context-aware feature transformation module is introduced in the generator to ensure that the generated feature encoding of either object or stuff is aware of other co-existing objects/stuff in the scene. Second, instead of feeding location-insensitive image features to the discriminator, we use the Gram matrix computed from the feature maps of the generated object images to preserve location-sensitive information, resulting in much enhanced object appearance. Extensive experiments show that the proposed method achieves state-of-the-art performance on the COCO-Thing-Stuff and Visual Genome benchmarks.

Sen He, Wentong Liao, Hamed R. Tavakoli et al.

Automatic captioning of images is a task that combines the challenges of image analysis and text generation. One important aspect in captioning is the notion of attention: How to decide what to describe and in which order. Inspired by the successes in text analysis and translation, previous work have proposed the \textit{transformer} architecture for image captioning. However, the structure between the \textit{semantic units} in images (usually the detected regions from object detection model) and sentences (each single word) is different. Limited work has been done to adapt the transformer's internal architecture to images. In this work, we introduce the \textbf{\textit{image transformer}}, which consists of a modified encoding transformer and an implicit decoding transformer, motivated by the relative spatial relationship between image regions. Our design widen the original transformer layer's inner architecture to adapt to the structure of images. With only regions feature as inputs, our model achieves new state-of-the-art performance on both MSCOCO offline and online testing benchmarks.

Sen He, Hamed R. Tavakoli, Ali Borji et al.

Recently, data-driven deep saliency models have achieved high performance and have outperformed classical saliency models, as demonstrated by results on datasets such as the MIT300 and SALICON. Yet, there remains a large gap between the performance of these models and the inter-human baseline. Some outstanding questions include what have these models learned, how and where they fail, and how they can be improved. This article attempts to answer these questions by analyzing the representations learned by individual neurons located at the intermediate layers of deep saliency models. To this end, we follow the steps of existing deep saliency models, that is borrowing a pre-trained model of object recognition to encode the visual features and learning a decoder to infer the saliency. We consider two cases when the encoder is used as a fixed feature extractor and when it is fine-tuned, and compare the inner representations of the network. To study how the learned representations depend on the task, we fine-tune the same network using the same image set but for two different tasks: saliency prediction versus scene classification. Our analyses reveal that: 1) some visual regions (e.g. head, text, symbol, vehicle) are already encoded within various layers of the network pre-trained for object recognition, 2) using modern datasets, we find that fine-tuning pre-trained models for saliency prediction makes them favor some categories (e.g. head) over some others (e.g. text), 3) although deep models of saliency outperform classical models on natural images, the converse is true for synthetic stimuli (e.g. pop-out search arrays), an evidence of significant difference between human and data-driven saliency models, and 4) we confirm that, after-fine tuning, the change in inner-representations is mostly due to the task and not the domain shift in the data.

Sen He, Dmitry Kangin, Yang Mi et al.

In this paper, we apply the attention mechanism to autonomous driving for steering angle prediction. We propose the first model, applying the recently introduced sparse attention mechanism to visual domain, as well as the aggregated extension for this model. We show the improvement of the proposed method, comparing to no attention as well as to different types of attention.

Sen He, Nicolas Pugeault

Saliency prediction is a well studied problem in computer vision. Early saliency models were based on low-level hand-crafted feature derived from insights gained in neuroscience and psychophysics. In the wake of deep learning breakthrough, a new cohort of models were proposed based on neural network architectures, allowing significantly higher gaze prediction than previous shallow models, on all metrics. However, most models treat the saliency prediction as a \textit{regression} problem, and accurate regression of high-dimensional data is known to be a hard problem. Furthermore, it is unclear that intermediate levels of saliency (ie, neither very high, nor very low) are meaningful: Something is either salient, or it is not. Drawing from those two observations, we reformulate the saliency prediction problem as a salient region \textit{segmentation} problem. We demonstrate that the reformulation allows for faster convergence than the classical regression problem, while performance is comparable to state-of-the-art. We also visualise the general features learned by the model, which are showed to be consistent with insights from psychophysics.

Sen He, Ali Borji, Yang Mi et al.

Deep convolutional neural networks have demonstrated high performances for fixation prediction in recent years. How they achieve this, however, is less explored and they remain to be black box models. Here, we attempt to shed light on the internal structure of deep saliency models and study what features they extract for fixation prediction. Specifically, we use a simple yet powerful architecture, consisting of only one CNN and a single resolution input, combined with a new loss function for pixel-wise fixation prediction during free viewing of natural scenes. We show that our simple method is on par or better than state-of-the-art complicated saliency models. Furthermore, we propose a method, related to saliency model evaluation metrics, to visualize deep models for fixation prediction. Our method reveals the inner representations of deep models for fixation prediction and provides evidence that saliency, as experienced by humans, is likely to involve high-level semantic knowledge in addition to low-level perceptual cues. Our results can be useful to measure the gap between current saliency models and the human inter-observer model and to build new models to close this gap.

Sen He, Nicolas Pugeault

Deep convolutional neural networks have achieved impressive performance on a broad range of problems, beating prior art on established benchmarks, but it often remains unclear what are the representations learnt by those systems and how they achieve such performance. This article examines the specific problem of saliency detection, where benchmarks are currently dominated by CNN-based approaches, and investigates the properties of the learnt representation by visualizing the artificial neurons' receptive fields. We demonstrate that fine tuning a pre-trained network on the saliency detection task lead to a profound transformation of the network's deeper layers. Moreover we argue that this transformation leads to the emergence of receptive fields conceptually similar to the centre-surround filters hypothesized by early research on visual saliency.