Research map / Video / Video-language understanding
Minerva-Ego: Spatiotemporal Hints for Egocentric Video Understanding
Arsha Nagrani, Jasper Uijlings, Shyamal Buch, Tobias Weyand, Sudheendra Vijayanarasimhan, Bo Hu, Ramin Mehran, David A Ross, Cordelia Schmid
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 |
|---|---|---|---|
| enable thinking (ablation) | Gemini 2.5 Flash · Minerva-Ego | MCQ accuracy (%): 31.7 -> 35.6 | Improved |
| increase frames from 64 to 1024 (ablation) | Gemini 2.5 Pro · Minerva-Ego | MCQ accuracy (%): 40.1 -> 49.8 | Improved |
| increase frames from 64 to 1024 (ablation) | Gemini 2.5 Flash · Minerva-Ego | MCQ accuracy (%): 31.7 -> 40.8 | Improved |
| provide visual frames instead of QA-only (ablation) | Gemini 2.5 Pro · Minerva-Ego | MCQ accuracy (%): 27.5 -> 40.1 | Improved |
| add temporal selection of frames (oracle) (ablation) | Gemini 2.5 Pro with 256 frames · Minerva-Ego | MCQ accuracy (%): 44.5% -> 47.4% | Improved |
| highlight objects with circles instead of masks or boxes (oracle) (ablation) | Gemini 2.5 Pro with 256 frames · Minerva-Ego | MCQ accuracy (%): 44.5% -> 45.9% | Improved |
| draw class names on frames (oracle) (ablation) | Gemini 2.5 Pro with 256 frames · Minerva-Ego | MCQ accuracy (%): 44.5% -> 47.1% | Improved |
| combine circles, class names and temporal selection (oracle) (ablation) | Gemini 2.5 Pro with 256 frames · Minerva-Ego | MCQ accuracy (%): 44.5% -> 50.3% | Improved |
| draw masks as highlight (oracle, with temporal selection) (ablation) | Gemini 2.5 Pro with 256 frames · Minerva-Ego | MCQ accuracy (%): 47.4% -> 47.3% | No clear effect |
| interleave frames with text boxes and classes instead of in-frame highlighting (oracle) (ablation) | Gemini 2.5 Pro with 256 frames · Minerva-Ego | MCQ accuracy (%): 50.3% -> 48.4% | Hurt |
| add temporal selection with OWLv2 predicted objects (ablation) | Gemini 2.5 Pro with 256 frames · Minerva-Ego | MCQ accuracy (%): 44.5% -> 46.5% | Improved |
| add class names with temporal selection (predicted objects) (ablation) | Gemini 2.5 Pro with 256 frames · Minerva-Ego | MCQ accuracy (%): 44.5% -> 47.0% | Improved |
Closest papers by meaning
- EgoThinker: Unveiling Egocentric Reasoning with Spatio-Temporal CoT (NeurIPS 2025 · similarity 0.94)
- EgoExoBench: A Benchmark for First- and Third-person View Video Understanding in MLLMs (NeurIPS 2025 · similarity 0.91)
- Omnia de EgoTempo: Benchmarking Temporal Understanding of Multi-Modal LLMs in Egocentric Videos (CVPR 2025 · similarity 0.91)
- Seeing Together: Multi-Robot Cooperative Egocentric Spatial Reasoning with Multimodal Large Language Models (NeurIPS 2026 · similarity 0.90)
- EgoProx: Evaluating MLLMs on Egocentric 3D Proximity Reasoning Across a Cognitive Hierarchy (CVPR 2026 · similarity 0.90)
- MMEgo: Towards Building Egocentric Multimodal LLMs for Video QA (ICLR 2025 · similarity 0.89)
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.
