Stories tagged “Computer Vision.”
24 stories
Researchers propose QFedPolyp, a communication- and inference-efficient federated learning framework for collaborative polyp segmentation. The framework combines quantization-aware training with low-precision model communication, reducing transmission cost and preserving …
Kimi Opens K3 Weight, Seeks Global Ambassadors, What Does Kimi Want This Time?
Kimi has opened the weight of K3, a 2.8 billion parameter hybrid expert model, and is seeking global ambassadors. The model has been released with a technical report and supporting infrastructure, including MoonEP, FlashKDA, and AgentEnv. Kimi wants to find people who hav…
This repository hosts code examples for a vast array of computer vision and AI topics, directly supporting articles on LearnOpenCV.com.
Hugging Face has open-sourced PyTorch Image Models, a repository of pre-trained vision transformer models, including the popular ViT and ConvNeXt models.
Researchers propose DocOCR-Eval, an annotation-free evaluation framework for automatic OCR assessment and selection. They conduct a systematic evaluation of text recognition performance across a diverse set of OCR engines and state-of-the-art MLLMs on multiple scanned doc…
NVIDIA has released Cosmos 3 Edge, a 4-billion-parameter open world model that helps robots and vision AI agents understand their surroundings, reason in real time, and generate robot actions on edge devices.
Researchers introduce qZACH-ViT, a quantization-aware extension of the zero-token ZACH-ViT backbone with recursive intrinsic patch-level class evidence. They also introduce Recursive Attribution-Stabilized Optimization (RASO), which norm-matches classification and attribu…
Streaming augmentations improve AI agents' imitation learning in streamed video games. They boost performance and robustness against visual artifacts and network lag, cutting data needs.
Researchers propose a new method for improving zero-shot image classification using pre-trained vision-language models. The approach, called class-aware zero-shot prompt reweighting, adjusts the weighting vector for each class label to capture the class-specific relevance…
Researchers have developed a quantum-hybrid solution for wildfire detection from satellite imagery, using a U-Net image segmentation model with a variational quantum circuit. The model outperformed classical U-Net results and showed potential for quantum machine learning …
Deep learning models face deployment challenges due to computational costs and energy demands. Researchers propose a lightweight training strategy that adapts normalization layers of the model to the new domain and decouples feature extraction from classifier optimization.
SenseTime has open-sourced SenseNova-Vision, a unified vision model designed to handle multiple computer vision tasks within a single system, eliminating the need for separate specialist models.
Is the Era of Embodied Intelligence's 'iPhone Moment' Finally Here?
Experts discuss the current state of embodied intelligence, a field that aims to bring artificial intelligence into the physical world. They touch on the challenges and opportunities in this space, including the need for more data, the development of more sophisticated mo…
Researchers investigate the issue of rare-label fairness in chest X-ray classification, where certain subgroups of patients are more likely to be misdiagnosed due to the long-tailed nature of the data. They propose a method to reduce the false negative rate in these subgr…
Ant Group's Robbyant unit has unveiled new AI vision models, LingBot-Depth 2.0 and LingBot-Vision, designed to improve robot perception of transparent objects and complex environments.
Ultralytics YOLOv5 is a fast, accurate, and easy-to-use computer vision model built on PyTorch, widely adopted for object detection, image segmentation, and classification tasks.
A new RANSAC variant eliminates the need for a user-supplied inlier scale parameter by analytically marginalizing the inlier scale, improving robustness and accuracy across diverse data conditions.
The RealSense SDK 2.0 is a cross-platform library for RealSense depth cameras, enabling depth and color streaming with calibration data.
Researchers developed an automated pipeline to systematically explore heterogeneous 4-Expert Mixture-of-Experts (MoE4) architectures, identifying optimal expert combinations and a significant coverage bias in their initial search.
This research diagnoses and repairs issues in learning-based single-shot fringe projection profilometry (FPP) for long-range applications, improving accuracy by architecturally removing reliance on shape priors.
A new approach to prefix caching in machine learning models allows for editable and composable notes, reducing latency and improving performance. The method enables the reuse of prefill across different contexts, making it more efficient and flexible.
The paper proposes a nonlinear concept-bottleneck method that aims to keep predictions explainable without relying on many concepts. It reports strong results, including in overhead-image object detection.
CAFD adds a concept-based signal from vision-language models to improve DNN fault detection while keeping computation practical.
The paper uses transcoders to trace how image inputs shape token generation in a VLM, and finds the method helps explain visual grounding and hallucinations.