Tagged: computer-vision

35 articles

Latest

Fine-Tuned CLIP Searches 12,000+ Satellite Images With Plain Text

Perceiver IO breaks the Transformer's quadratic bottleneck for images

Fine-tune Google’s ViT on bean leaves in under 100 lines of Python

SegFormer Fine-Tuning on a Sidewalk Dataset Shows Why Custom Data Beats Generic

Graphcore and Hugging Face drop 10 IPU-ready transformer models

Hugging Face Datasets Adds Audio and Vision Support, Kills Loading Scripts

Hugging Face AutoTrain adds image classification, hits 84% accuracy with zero code

CLIPSeg brings zero-shot segmentation to any object, no retraining needed

OneFormer unifies image segmentation, beats Mask2Former on all 3 tasks

Hugging Face now supports 8 core vision tasks and 100k+ models

Ex-Meta researchers release Isaac 0.5 to make warehouse robots actually think

Sonar-vision combo lets underwater robots see through sediment clouds

Ukraine hands UK 5 million labeled combat images to train AI weapons

Text-to-video models can't keep up with images — here's what's blocking them

The Hidden Flaws in Object Detection Benchmarks: What Your AP Score Isn't Telling You

No AI model can truly see: best score on new perception test is 59.7%

PSNR and SSIM reward blurry erasures — Xiaomi's PROVE metric finally catches it

ByteDance's SeedRealtime AI interrupted a bad espresso shot — and that changes everything

PixelRAG ditches HTML parsing, retrieves documents as screenshots with 7-query Recall@k eval

LingBot-Map Demo: 3D Scene Reconstruction Tunes Itself to Your GPU's VRAM

A single doctored photo can now fool AI from any angle

DeepMind: Video generators are the world models CV has been waiting for

OpenAI Puts Neural Networks Under a Microscope

GPT-2 trained on pixels rivals top CNNs without labels

CLIP drops the training wheels: Zero-shot vision matches ResNet-50

OpenAI finds CLIP has 'Halle Berry' neurons like the human brain

OpenAI lets you fine-tune GPT-4o with images, and yes, it's a big deal

Grab fine-tunes GPT-4o vision on just 100 images, maps SE Asia

A 0.6B model just beat SAM 3 on grounding—here's the simple trick that made it work

Hugging Face trains a 2.2B model to see GUIs, hitting 65% on ScreenSpot

Google's SigLIP 2 adds a decoder and self-distillation to build vision encoders that actually know where things are

Hugging Face bridges its entire pipeline to timm's 200K daily users via TimmWrapper

InstantMesh now converts raw 3D meshes into textured models in 10 seconds

Google's PaliGemma drops: 3 resolutions, 3 precisions, zero conversation

Pollen Robotics drops open-source vision stack that finds any object in 3D