Yuanbo Lyu, Changjian Pan, Saining Zhang et al.
High-quality articulated assets are essential for embodied perception, simulation, and manipulation, yet constructing them at scale remains difficult.High-quality articulated assets are essential for embodied perception, simulation, and manipulation, yet constructing them at scale remains difficult. Manual annotation and procedural modeling require substantial human effort.
Khai Le-Duc, Nguyen Kim Hai Bui, An Ngo et al.
Vision-language models can get individual judgments right while struggling to recover and consistently use the exact properties of a 3D object.Vision-language models can get individual judgments right while struggling to recover and consistently use the exact properties of a 3D object.
Jiawei Chen
Vision-language models (VLMs) are increasingly prompted with personas, such as "You have low vision", to stand in for people in user studies.Vision-language models (VLMs) are increasingly prompted with personas, such as "You have low vision", to stand in for people in user studies. We ask whether a persona makes a model see like the person it describes.
Rishav Pramanik, Marco Pedersoli, Zhaozheng Yin
Autoregressive (AR) image generators are powerful in principle but inflexible in practice: training with a fixed raster-scan order hardcodes a single factorization of the…Autoregressive (AR) image generators are powerful in principle but inflexible in practice: training with a fixed raster-scan order hardcodes a single factorization of the image distribution into the model, leaving it unable to…
Ankit Khedia, Astitva Jaiswal, Chaitanya Prasad et al.
Video captioning models are increasingly evaluated with metrics designed for single-sentence image captions.Video captioning models are increasingly evaluated with metrics designed for single-sentence image captions. We test whether those metrics recover human judgement on dense, multi-window video description, and find that they do…
Yang-Che Tseng, Erqun Dong, Jeong-gi Kwak et al.
While single-view novel view synthesis methods have been dramatically enhanced recently, how to quantify their performance has yet to be standardized.While single-view novel view synthesis methods have been dramatically enhanced recently, how to quantify their performance has yet to be standardized.
JianChong Guo, YuTing Wan, Kefan Li et al.
Building damage assessment (BDA) from very-high-resolution imagery supports disaster response, and deep models now achieve high overall accuracy on public benchmarks.Building damage assessment (BDA) from very-high-resolution imagery supports disaster response, and deep models now achieve high overall accuracy on public benchmarks.
In LOLv2-Real, a test of low-light photo enhancement, reference images were 14% darker in the test set than in the training set.In LOLv2-Real, a test of low-light photo enhancement, reference images were 14% darker in the test set than in the training set. An audit of LOL-v1 and LOLv2-Real, two standard tests of methods that brighten dark photos, found…
Kang He, Wenshuo Peng, Chuanhao Li et al.
Recent advances in interactive world models have made it possible to generate controllable, dynamic environments from action inputs.Recent advances in interactive world models have made it possible to generate controllable, dynamic environments from action inputs. Yet current models struggle to maintain coherence beyond roughly one minute of rollout, a…
Vansh Bhatia, Mehdi Rezagholizadeh, Parsa Ashrafi Fashi et al.
Final-answer accuracy reports whether a vision-language model was right, but not whether its answer was grounded in the image, whether its stated reasoning reflects that…Final-answer accuracy reports whether a vision-language model was right, but not whether its answer was grounded in the image, whether its stated reasoning reflects that evidence, or what the answer cost to produce.
Tengjiao Sun, Pengcheng Fang, Xiaoyu Zhan et al.
Text-to-motion models generate motion only for the skeletons they were trained on, while a new character arrives as a skeleton file and nothing else.Text-to-motion models generate motion only for the skeletons they were trained on, while a new character arrives as a skeleton file and nothing else.
Patryk Bartkowiak, Dominik Michels, Sören Pirk et al.
Distilled text-to-image models generate images quickly, but classifiers trained on those images can perform substantially worse than classifiers trained on images from…Distilled text-to-image models generate images quickly, but classifiers trained on those images can perform substantially worse than classifiers trained on images from their slower base models.
Jialv Zou, Xiaotong Li, Zhengrong Yue et al.
Discrete visual tokens represent vision in a discrete form analogous to language, offering a native interface for multimodal modeling under a shared next-token prediction…Discrete visual tokens represent vision in a discrete form analogous to language, offering a native interface for multimodal modeling under a shared next-token prediction objective.
Wendi Ying, Daochang Liu, Leonid Sigal et al.
Existing evaluations of text-conditioned 3D indoor scene generation largely emphasize semantic fidelity and physical validity at rest, leaving interaction-time physical…Existing evaluations of text-conditioned 3D indoor scene generation largely emphasize semantic fidelity and physical validity at rest, leaving interaction-time physical reliability insufficiently assessed.
Xiaokang Yuan, Binhui Yao, Luwei Yang et al.
Leading GUI grounders score 93 to 96% on ScreenSpot-v2, and reported gaps between new and prior models are about a point.Leading GUI grounders score 93 to 96% on ScreenSpot-v2, and reported gaps between new and prior models are about a point. At this level the useful question is what the remaining errors are.
Lior Ziv, Mahmood Sharif
While few-step distilled text-to-image models have gained traction due to their fast generation times, the speed-up implications on their unsafe generation rates remain…While few-step distilled text-to-image models have gained traction due to their fast generation times, the speed-up implications on their unsafe generation rates remain unknown.
Truman Hickok
Low-cost arms such as the SO-101 have become the most common entry point to robot learning, yet no released policy performs spatially constrained, language-conditioned…Low-cost arms such as the SO-101 have become the most common entry point to robot learning, yet no released policy performs spatially constrained, language-conditioned tasks on them out of the box, and existing SO-101 benchmarks…
Bingliang Zhang, Wenda Chu, Yizhuo Li et al.
We present Scalable Pixel-anchored End-to-end Diffusion (SpeeDiff), a latent diffusion method that jointly trains the VAE and the diffusion model from scratch.We present Scalable Pixel-anchored End-to-end Diffusion (SpeeDiff), a latent diffusion method that jointly trains the VAE and the diffusion model from scratch.
Muxi Diao, Wenbin Xu, Lele Yang et al.
Diffusion video models produce striking clips, but they cannot reliably render an exact formula, a correct chart, a rule-consistent game or narration synchronized to…Diffusion video models produce striking clips, but they cannot reliably render an exact formula, a correct chart, a rule-consistent game or narration synchronized to on-screen events.
Chao-Yeh Chen, Tiffany Chen, Lin Duan et al.
We introduce EgoHand, a monocular 3D hand reconstruction model that achieves state-of-the-art performance on egocentric video.We introduce EgoHand, a monocular 3D hand reconstruction model that achieves state-of-the-art performance on egocentric video.
Minh Pham-Dinh, Md Farhan Ishmam, Hernan Yupanqui et al.
Browser agents do not just need more practice. They need to practice the right things. We introduce MiniWeb, a realistic miniature of the web comprising 65 interconnected…Browser agents do not just need more practice. They need to practice the right things. We introduce MiniWeb, a realistic miniature of the web comprising 65 interconnected websites across 15 domains, where agents navigate, search…
Diogo J. Paulo, Hugo Proença, João C. Neves
Face morphing attacks, in which the biometric traits of two or more identities are blended into a single face image, pose a severe threat to identity document issuance…Face morphing attacks, in which the biometric traits of two or more identities are blended into a single face image, pose a severe threat to identity document issuance and automated border control.
Seyedarmin Azizi, Parsa Razmara, Sajjad Ghiasvand et al.
Text-steerable visual encoders condition a frozen Vision Transformer on a natural-language prompt and point at the matching image region, producing prompt-conditioned…Text-steerable visual encoders condition a frozen Vision Transformer on a natural-language prompt and point at the matching image region, producing prompt-conditioned heatmaps and features.
Across seven medical datasets, detectors with random numerical replacements for text tracked those using correct descriptions when the distance between reference points…Across seven medical datasets, detectors with random numerical replacements for text tracked those using correct descriptions when the distance between reference points matched.
Yuesheng Huang, Hoi Fan Au, Jiaqi Hou et al.
Localized AI manipulations in long-form videos often affect brief events within otherwise authentic content, making video-level classification insufficient for forensic…Localized AI manipulations in long-form videos often affect brief events within otherwise authentic content, making video-level classification insufficient for forensic analysis.
StreamEval Team
We introduce StreamEval with four key features: (1) First interactive streaming video generation benchmark. (2) Fully open source.We introduce StreamEval with four key features: (1) First interactive streaming video generation benchmark. (2) Fully open source. Benchmark data, evaluation code, protocols, and scoring rules are publicly available.
Panagiotis Koromilas, Eleanna Vali, Diogenis Antonopoulos et al.
Location encoders map geographic coordinates to high-dimensional descriptors through contrastive alignment with paired Earth Observation imagery.Location encoders map geographic coordinates to high-dimensional descriptors through contrastive alignment with paired Earth Observation imagery.
Zhiran Yan, Savankumar Bhanderi, Sourabh Lolge et al.
Recent advances in 3D semantic segmentation have significantly improved scene understanding in autonomous driving.Recent advances in 3D semantic segmentation have significantly improved scene understanding in autonomous driving. However, existing methods are typically developed under a closed-set assumption and evaluated under standard…
Gibran Benitez-Garcia, Hiroki Takahashi
Hand landmarks serve as input for gesture recognition, sign language understanding, and touchless human-computer interaction.Hand landmarks serve as input for gesture recognition, sign language understanding, and touchless human-computer interaction.
Yanhao Jia, Yi Zhu, Yiran Zhong et al.
Humans efficiently process multimodal streams spanning hours to years, building persistent, fine-grained spatiotemporal memories that support reasoning about and…Humans efficiently process multimodal streams spanning hours to years, building persistent, fine-grained spatiotemporal memories that support reasoning about and predicting long-horizon events.
Xiangzhong Ye, Zhongshen Li, Shen You et al.
Fixed-window Vision Transformer (ViT) decoders run at deployment resolution as overlapping tiles: every token in an overlap band is processed by several full decoder…Fixed-window Vision Transformer (ViT) decoders run at deployment resolution as overlapping tiles: every token in an overlap band is processed by several full decoder instances.
An audit found approximately 10.5% of 400,000 documents had erroneous labels, and 35.2% of the test set had duplicates or near-duplicates in the training set.An audit found approximately 10.5% of 400,000 documents had erroneous labels, and 35.2% of the test set had duplicates or near-duplicates in the training set.
Bang-Shien Chen, Caitrin E. Eaton, David M. Rosen
Estimating a camera pose from 3D-2D correspondences - Perspective-n-Point - is one of the most fundamental problems in computer vision.Estimating a camera pose from 3D-2D correspondences - Perspective-n-Point - is one of the most fundamental problems in computer vision. Existing methods either solve a minimal problem or replace the objective with an easier one.
Kesav Kumar Jayakumar, Karthigeyan Thilak
We set out to reproduce a recent hallucination-mitigation result on the POPE benchmark. The LLaVA-1.5-7B baseline we measured, 82.21 F1 on the adversarial split, sat 1.21…We set out to reproduce a recent hallucination-mitigation result on the POPE benchmark. The LLaVA-1.5-7B baseline we measured, 82.21 F1 on the adversarial split, sat 1.21 to 8.67 points above the adversarial baselines we planned…
Advait Gupta, Sachin Konan, Tom Yin et al.
Vision-language models (VLMs) can perform a broad range of tasks across diverse settings, yet their performance in agentic contexts remains poorly understood.Vision-language models (VLMs) can perform a broad range of tasks across diverse settings, yet their performance in agentic contexts remains poorly understood.
Di Yang, Xiaochang Li, Muyang Liu et al.
Many questions about a video concern one physical object. Where does the ball end up after the cups are shuffled? Which card is the queen? How often was the mug covered?Many questions about a video concern one physical object. Where does the ball end up after the cups are shuffled? Which card is the queen? How often was the mug covered?
Nimrod Shabtay, Asaf Yehudai, Avihu Dekel et al.
Vision-language models (VLMs) struggle with in context learning compared to their textual counterparts.Vision-language models (VLMs) struggle with in context learning compared to their textual counterparts. Our preliminary study shows that VLMs rely on textual shortcuts and underutilize the visual demonstrations.
Keigo Watanabe, Hiroshi Kera, Kazuhiko Kawamoto
Low-rank adaptation (LoRA) is usually tied to the pretrained backbone on which it is learned. We ask whether LoRAs trained independently for the same task on different…Low-rank adaptation (LoRA) is usually tied to the pretrained backbone on which it is learned. We ask whether LoRAs trained independently for the same task on different pretrained backbones can be summed and applied unchanged to…
Seemandhar Jain, Manmohan Chandraker
Reproducing a published 3DGS paper takes weeks of expert engineering before any extension can be proposed.Reproducing a published 3DGS paper takes weeks of expert engineering before any extension can be proposed.
Nvidia, :, Sérgio Agostinho et al.
We introduce Foundation4D, a unified feed-forward model for geometry estimation, 3D detection, semantic occupancy, and Gaussian reconstruction from multi-camera driving…We introduce Foundation4D, a unified feed-forward model for geometry estimation, 3D detection, semantic occupancy, and Gaussian reconstruction from multi-camera driving video.