Research map / 3D vision / 3D generation
Not all Views are Created Equal: Analyzing Viewpoint Instabilities in Vision Foundation Models
Mateusz Michalkiewicz, Sheena Bai, Mahsa Baktashmotlagh, Varun Jampani, Guha Balakrishnan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| viewpoint is unstable (other) rather than stable | CLIP zero-shot · ABO | Acc@5: 68.74 -> 42.56 | Hurt |
| viewpoint is accidental rather than stable | CLIP zero-shot · ABO | Acc@5: 68.74 -> 3.00 | Hurt |
| viewpoint is accidental rather than stable | CLIP zero-shot · CO3D | Acc@1: 78.58 -> 48.36 | Hurt |
| viewpoint is other rather than stable | CLIP zero-shot · CO3D | Acc@1: 78.58 -> 68.45 | Hurt |
| apply OmniView-tuned OpenCLIP instead of CLIP on other viewpoints | CLIP · ABO | Acc@5: 42.56 -> 56.13 | Improved |
| apply OmniView-tuned OpenCLIP instead of CLIP on other viewpoints | CLIP · CO3D | Acc@1: 68.45 -> 77.92 | Improved |
| apply OmniView-tuned OpenCLIP on accidental viewpoints | CLIP · ABO | Acc@5: 3.00 -> 3.34 | No clear effect |
| fine-tune CLIP for multiview equivariance (CLIP-FT) | CLIP · ABO | Acc@5: 42.56 -> 43.17 | No clear effect |
| fine-tune CLIP for multiview equivariance (CLIP-FT) | CLIP · CO3D | Acc@1: 68.45 -> 69.52 | No clear effect |
| viewpoint is accidental rather than stable | Stable Fast 3D reconstruction · ABO | F1@0.5: 82.86% -> 36.67% | Hurt |
| viewpoint is accidental rather than stable | LLaVA-1.5 VQA · ABO | F1@0.5: 56.42% -> 29.86% | Hurt |
| viewpoint is accidental rather than stable | CLIP linear probe · ABO | accuracy: 89.62% -> 63.19% | Hurt |
Closest papers by meaning
- Probing the 3D Awareness of Visual Foundation Models (CVPR 2024 · similarity 0.87)
- Evaluating Multiview Object Consistency in Humans and Image Models (NeurIPS 2024 · similarity 0.85)
- Seeing from Another Perspective: Evaluating Multi-View Understanding in MLLMs (AAAI 2026 · similarity 0.85)
- MMPerspective: Do MLLMs Understand Perspective? A Comprehensive Benchmark for Perspective Perception, Reasoning, and Robustness (NeurIPS 2025 · similarity 0.85)
- Directional Confusions Reveal Divergent Inductive Biases Through Rate-Distortion Geometry in Human and Machine Vision (NeurIPS 2026 · similarity 0.84)
- Can 4D Foundation Models Remember? (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
