Computer Vision

Vision transformers surprisingly struggle with visual relations; this study reveals ViTs use distinct perceptual and relational processing stages to solve same/different tasks, highlighting a previous…

Beyond Euclidean spaces, Dual-Space Representation Learning (DSRL) enhances weakly supervised video violence detection by cleverly integrating Euclidean and hyperbolic geometries for superior discrimi…

CPDTrack: Human-like Visual Search Boosts Object Tracking!

BetterDepth: A plug-and-play diffusion refiner boosts zero-shot monocular depth estimation by adding fine details while preserving accurate geometry.

O-BELM, a novel diffusion model sampler, achieves mathematically exact inversion with superior sampling quality, offering a new gold standard for diffusion model applications.

Bayesian-PEFT boosts vision model accuracy and confidence in few-shot learning by integrating Bayesian components into PEFT, solving the underconfidence problem.

AverNet: All-in-one video restoration defying time-varying unknown degradations.

GNNs automate multi-dataset semantic segmentation label unification, improving model training efficiency and performance by resolving conflicts across label spaces.

AUCSeg tackles pixel-level long-tail semantic segmentation by introducing an AUC-oriented loss function and a Tail-Classes Memory Bank to efficiently manage memory and improve performance on imbalance…

Boosting Vision Transformer’s transferability in cross-domain few-shot learning is achieved by a simple yet effective method: strategically adjusting attention temperature to remedy ineffective target…

ZoDiac: a novel image watermarking framework leveraging pre-trained stable diffusion models for robust, invisible watermarks resistant to state-of-the-art attacks.

APM: Asynchronous Perception Machine, a computationally-efficient architecture for test-time training (TTT), processes image patches asynchronously, encoding semantic awareness without pre-training, a…

AsyncDiff accelerates diffusion model inference by 2.8x using asynchronous denoising and model parallelism, maintaining near-perfect image quality.

Hypergraph-Based Assembly Fuzzy Representation (HAFR) excels at open-set 3D object retrieval by using part-level shapes and fuzzy representations to overcome challenges posed by unseen object categori…

AsCAN, a novel hybrid architecture, achieves superior efficiency and performance in image recognition and generation by asymmetrically combining convolutional and transformer blocks.

Unsupervised Articulated Object Modeling using Conditional View Synthesis learns pose and part segmentation from only two object observations, achieving significantly better performance than previous …

ART: A novel automatic red-teaming framework reveals safety vulnerabilities in popular text-to-image models by identifying unsafe outputs even from seemingly harmless prompts.

Focusing on nuclear morphology improves out-of-domain generalization in cancer classification from histopathology images by leveraging nuclear segmentation masks during training.

Large-scale dataset distillation can be achieved with significantly less soft labels by using class-wise supervision during image synthesis, enabling simple random label pruning and enhancing model ac…

Boosting image generation: Applying guidance selectively during diffusion model sampling drastically enhances image quality and inference speed, achieving state-of-the-art results.