Research map / Video / Video-language understanding
Learning Egocentric Cues from Exocentric Video using Privileged Egocentric Supervision
Dominick Reilly, Manish Govind, Le Xue, Srijan Das
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 |
|---|---|---|---|
| add Ego2Exo sequence distillation (ablation) | Exo-only VideoLLaMA3 · Ego-in-Exo Perception | avg accuracy (%): 65.7 -> 68.5 | Improved |
| add Ego Adaptive Visual Tokens (ablation) | Exo-only VideoLLaMA3 · Ego-in-Exo Perception | avg accuracy (%): 65.7 -> 72.2 | Improved |
| add both distillation and tokens (ablation) | Exo-only VideoLLaMA3 · Ego-in-Exo Perception | avg accuracy (%): 65.7 -> 74.2 | Improved |
| add ego videos to training data without distillation (ablation) | Exo-only VideoLLaMA3 · Ego-in-Exo Perception | avg accuracy (%): 65.7 -> 66.0 | No clear effect |
| use cross-attention to vision encoder features instead of vision-encoder self-attention for tokens (ablation) | Ego2ExoVLM · Ego-in-Exo Perception | avg accuracy (%): 69.9 -> 74.2 | Improved |
| train tokens jointly instead of initializing from ego-pretrained cross-attention (ablation) | Ego2ExoVLM · Ego-in-Exo Perception | avg accuracy (%): 73.4 -> 74.2 | Improved |
| use language-level sequence distillation instead of visual feature distillation (ablation) | Ego2ExoVLM · Ego-in-Exo Perception | avg accuracy (%): 67.8 -> 74.2 | Improved |
| use ego-to-exo language distillation instead of Exo2EgoVLM (ablation) | Ego2ExoVLM · Ego-in-Exo Perception | avg accuracy (%): 70.3 -> 74.2 | Improved |
| train on ego-pretrained LLAVIDAL-style token transfer instead (ablation) | Ego2ExoVLM · Ego-in-Exo Perception | avg accuracy (%): 65.4 -> 74.2 | Improved |
| train with ego-exo distillation instead of exo-only fine-tuning on same data | VideoLLaMA3 · Ego-in-Exo Perception | avg accuracy (%): 65.7 -> 74.2 | Improved |
| add Ego2ExoVLM to VideoLLaMA3 | VideoLLaMA3 · Ego-in-Exo Perception | avg accuracy (%): 71.2 -> 74.2 | Improved |
| add Ego2ExoVLM to weaker CLIP/Mistral backbone | CLIP-Large / Mistral v0.2 · Ego-in-Exo Perception | avg accuracy (%): 37.0 -> 45.7 | Improved |
Closest papers by meaning
- Exo2Ego: Exocentric Knowledge Guided MLLM for Egocentric Video Understanding (AAAI 2026 · similarity 0.93)
- EgoExoBench: A Benchmark for First- and Third-person View Video Understanding in MLLMs (NeurIPS 2025 · similarity 0.91)
- EgoThinker: Unveiling Egocentric Reasoning with Spatio-Temporal CoT (NeurIPS 2025 · similarity 0.90)
- Exocentric-to-Egocentric Video Generation (NeurIPS 2024 · similarity 0.89)
- Retrieval-Augmented Egocentric Video Captioning (CVPR 2024 · similarity 0.89)
- Test-Time Adaptation for Combating Missing Modalities in Egocentric Videos (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.
