Hard Vision, Easy Vision: What GPT-6 Astra Reveals Across Computer Vision

arXiv:2609.35718 · cs.CV · Submitted 2026-09-28 · Read on arXiv

cs.CV

Submitted: 2026-09-28

Updated: 2026-09-28

Terminology

• Grounding — 40× in this paper · explained in 3 episodes
• Benchmark — 30× in this paper · explained in 2 episodes
• Visual Grounding — 12× in this paper · explained in 2 episodes
• Accuracy — 9× in this paper · explained in 1 episode
• Conference on Computer Vision and Pattern Recognition — 8× in this paper
• Consistency — 5× in this paper · explained in 3 episodes
• Benchmarking — 4× in this paper · explained in 3 episodes
• Fidelity — 4× in this paper · explained in 2 episodes
• Foundation Model — 4× in this paper · explained in 3 episodes
• Temporal consistency — 4× in this paper · explained in 2 episodes
• Adam — 3× in this paper · explained in 2 episodes
• Foundation Models — 3× in this paper · explained in 7 episodes
• Transformer — 3× in this paper · explained in 2 episodes
• Decoupling — 2× in this paper · explained in 2 episodes
• Trace — 2× in this paper · explained in 2 episodes
• Adaptation — 1× in this paper · explained in 1 episode
• Alignment — 1× in this paper · explained in 1 episode
• Core — 1× in this paper · explained in 2 episodes
• Hallucination — 1× in this paper · explained in 3 episodes
• Instance Segmentation — 1× in this paper · explained in 3 episodes
• Instruction Tuning — 1× in this paper · explained in 2 episodes
• Multimodal Reasoning — 1× in this paper · explained in 5 episodes
• Point Clouds — 1× in this paper · explained in 2 episodes
• Reconstruction Error — 1× in this paper · explained in 2 episodes
• Regularization — 1× in this paper · explained in 3 episodes
• Semantic Understanding — 1× in this paper · explained in 2 episodes
• Task Planning — 1× in this paper · explained in 2 episodes
• Temporal Reasoning — 1× in this paper · explained in 2 episodes

Sources

• V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning
• M3D: Advancing 3D Medical Image Analysis with Multi-Modal Large Language Models
• UniPercept: Towards Unified Perceptual-Level Image Understanding across Aesthetics, Quality, Structure, and Texture
• Boogu-Image-0.1: Boosting Open Agentic Multimodal Generation via Understanding under a Minimal Budget
• BabyVision: Visual Reasoning Beyond Language
• Retrieval-Based Interleaved Visual Chain-of-Thought in Real-World Driving Scenarios
• TRACE: Temporal Grounding Video LLM via Causal Event Modeling
• Human Cognitive Benchmarks Reveal Foundational Visual Gaps in MLLMs
• WildDet3D: Scaling Promptable 3D Detection in the Wild
• OpenAI o1 System Card
• GenEval 2: Addressing Benchmark Drift in Text-to-Image Evaluation
• Depth Anything 3: Recovering the Visual Space from Any Views
• PerceptionBench: Evaluating Atomic Visual Perception in Multimodal Large Language Models
• VLRS-Bench: A Vision-Language Reasoning Benchmark for Remote Sensing
• DINOv2: Learning Robust Visual Features without Supervision
• The 2017 DAVIS Challenge on Video Object Segmentation
• ZeroBench: An Impossible Visual Benchmark for Contemporary Large Multimodal Models
• VisualPuzzles: Decoupling Multimodal Reasoning Evaluation from Domain Knowledge
• CountQA: How Well Do MLLMs Count in the Wild?
• Gemini: A Family of Highly Capable Multimodal Models

Related papers

• Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
• AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
• Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
• MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
• TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
• A Survey on Efficient Vision-Language-Action Models