Computer Vision news and analysis on Web Pulse: 4294 curated articles tracking the latest Computer Vision developments, tools, and research, updated continuously from vetted sources.
Tencent released Hy-Image-3.5-Preview, a unified image generation and editing model that handles text-to-image, image-to-image, and multi-turn editing through one endpoint, accepting up to 20 reference images per request…
LightOn released LightOnOCR 3, an Apache 2.0 OCR model family in 0.8B, 1B and 4B sizes that returns a bounding box for every extracted block so chart values and table cells can be traced back to their exact location on t…
A developer built Touch Grass Bingo, an offline nature bingo web app that uses the open-weight CLIP ViT-B/32 vision model, 8-bit quantized to about 150 MB, running entirely in the phone's browser via Transformers.js on O…
A developer built Plain Sight Walk, a browser-based walking game that turns outdoor scavenger hunts into a photo carousel, using Gemma 2B via WebLLM on WebGPU to generate riddles and CLIP ViT-B/32 via transformers.js to …
A developer built HomeDesign, an AI web app that generates interior, exterior, and floor-plan design ideas from photos of a user's existing space. Its Exterior Paint Visualizer deliberately limits each generation to one …
VideoDB's Action Segmentation + Search pipeline returned a relevant moment in its top 10 results for 40.39% of 11,369 text queries on a 100-hour benchmark of 2,835 egocentric videos, about 5 points ahead of AWS Nova (35.…
A feasibility study of 150 patients at increased risk of basal cell carcinoma found that AI-assisted line-field confocal optical coherence tomography (LC-OCT) imaging identified histologically confirmed subclinical tumor…
A developer built Grass Quest, a Flask web app that uses Open-Meteo weather data and a locally run Llama 3.2 (3B) model via Ollama to generate one short, weather-appropriate outdoor mission per day, then scores user-subm…
A developer built TouchGrass AI, an open-source web app that uses open-weight models via Hugging Face Inference Providers to generate personalized outdoor missions and analyze user-uploaded nature photos, aiming to push …
Immich released version 3.3, adding shared people management, birthday memories, automatic stacking of edited photos, and machine learning optimizations, the open-source self-hosted photo and video management project ann…
A developer built Trailnotes, an open-source tool that turns a folder of walk photos into a field journal by running the open-weight vision model qwen2.5vl:7b locally via Ollama on an RTX 5060 laptop GPU, adding time and…
A developer built TrailMate AI, a browser-based outdoor exploration companion written in HTML, CSS, and vanilla JavaScript that gamifies time outside with adventure challenges, XP rewards, achievements, and a "Phone Down…
A developer built SeeMore AI, an observation-training tool that uses AI to generate indirect clues about photos of everyday surroundings instead of identifying objects outright, prompting users to look at the real world …
Researchers released a camera-native stereo VR180 dataset of 1,211 samples captured with two Blackmagic URSA Cine Immersive cameras, comprising 636 stereo video clips (2,220.8 seconds, mostly 90 fps) and 575 stereo still…
Researchers proposed SAVER, a dense-to-sparse post-training framework that trains video-language models to reason from limited temporal evidence, according to a new arXiv paper (2610.10893v1). Trained on only 1,250 rando…
Researchers proposed VICO, a co-evolutionary framework that jointly trains a vision-language model actor and an Environment-as-Rewriter that edits verifiable image-side structures such as scene graphs, chart tables, and …
SPLIT-RL, a staged post-training approach that trains visual reasoning and language reasoning in disjoint phases, improves average accuracy over GRPO by 1.4 to 6.1 points across Qwen3-VL models from 2B to 30B-A3B and Int…
A paper posted to arXiv (arXiv:2610.10959v1) introduces TopoGPU, a GPU streaming pipeline that computes persistence diagrams of cubical complexes induced by 2D and 3D images, achieving an average end-to-end speedup of 53…
Fluid-Gen-Zero, a training-free framework from researchers behind the arXiv:2610.10984v1 paper, grounds pretrained video generators in physics by delegating motion dynamics to a simulator while preserving the appearance …
A new arXiv paper (2610.10666v1) offers a theoretical explanation for why adversarially-trained neural networks produce sparse gradient saliency maps, proving that for two-layer ReLU networks, minimizers converge to a Ba…