Vision

Image and video understanding, multimodal perception, detection, segmentation, 3D understanding

30 articles

All Agents Language models Reinforcement learning Vision Generative media Speech & audio Robotics & embodied AI Safety & security Evaluation & benchmarks Efficiency & systems Code & software AI for science & health Theory

Viewgrid list compact

Training-free visual memory framework compresses video tokens while preserving event coherence

Decoupled temporal axis yields motion-prioritized video representations efficiently

Pruning visual tokens and decoder sublayers cuts multimodal model computation

Self-distillation helps vision-language models reason with far fewer visual tokens

Preserving viewpoint evidence improves text-guided 3D scene segmentation

Interleaved table crops improve reasoning over dense visual layouts

Fine-Grained Rewards Reduce Vision-Language Hallucinations Without Silencing Valid Detail

Synthetic multi-view supervision enables 4D rendering from static camera video

Steering hidden states stabilizes vision-language reasoning against subtle visual changes

Two same-image global views drive semantic structure in DINO models

Asymmetric certainty gains from optimization hinder multimodal classification

Parallel tile inspection helps models find details in large images

Forensic tool use improves explainable zero-shot image forgery detection

Visual image rearrangement helps models when exact evidence matters most

Shared scene graphs connect interaction understanding with grounded robot planning

Parallel tile inspection and adaptive zoom improve high-resolution visual search

Grounding long-video memories in visual identity lets QA systems track objects across days

Multi-agent framework generates spatial QA data via code execution

Token-aware latent refinement improves vision-language reasoning at inference time

Geometric tools let a small vision-language model measure 3D scenes

Hierarchical implicit models improve compact brain MRI segmentation across sites

End-to-end model writes 3D scene graphs directly from point clouds or Gaussian splats

Latent semantics from LLMs improve sentiment analysis with missing data

Agentic document QA pays off mainly with stronger vision-language models

Branched YOLOv2 with geometric features improves traffic sign recognition speed and accuracy.

4D foundation models lack robust visual memory for out-of-view objects

Streaming video memory works better when internalized as evolving latent tokens

Self-improving context programs help frozen models understand long videos

Multi-agent framework iteratively verifies interior layouts against architectural constraints

Open-source image model rivals closed systems on minimal budget