Research map / Video / Action recognition
Egocentric Activity Recognition and Localization on a 3D Map
Miao Liu, Lingni Ma, Kiran Somasundaram, Yin Li, Kristen Grauman, James M. Rehg, Chao Li
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 HVR 3D environment context to I3D-Res50 | I3D-Res50 · Ego4D subset (seen environments) | mean class accuracy: 4.2% | Improved |
| add HVR 3D environment context to I3D-Res50 | I3D-Res50 · Ego4D subset (seen environments) 3D localization | F1: 9.3% | Improved |
| use HVR semantic occupancy map instead of semantic-only voxel (ablation) | HVR · Ego4D subset (seen environments) | mean class accuracy: 39.23 -> 41.64 | Improved |
| use children voxel resolution M=4 instead of M=2 (ablation) | HVR · Ego4D subset (seen environments) | mean class accuracy: 39.04 -> 41.64 | Improved |
| increase voxel resolution from M=4 to M=8 (ablation) | HVR · Ego4D subset (seen environments) | mean class accuracy: 41.64 -> 40.06 | Hurt |
| increase voxel resolution from M=4 to M=8 (ablation) | HVR · Ego4D subset (seen environments) 3D localization | F1: 22.73 -> 22.96 | No clear effect |
| select local environment features with predicted 3D location instead of global features (ablation) | HVR · Ego4D subset (seen environments) | mean class accuracy: 35.99 -> 41.64 | Improved |
| fuse global environment features directly into I3D-Res50 (ablation) | I3D-Res50 · Ego4D subset (seen environments) | mean class accuracy: 37.48 -> 35.99 | Hurt |
| add Gumbel-Softmax stochastic location units (ablation) | HVR · Ego4D subset (seen environments) | mean class accuracy: 39.37 -> 41.64 | Improved |
| add Gumbel-Softmax stochastic location units (ablation) | HVR · Ego4D subset (seen environments) 3D localization | F1: 19.71 -> 22.73 | Improved |
| add HVR to I3D-Res50 on unseen environments | I3D-Res50 · Ego4D subset (unseen environments) | mean class accuracy: 29.24 -> 31.55 | Improved |
| add HVR to I3D-Res50 on unseen environments | I3D-Res50 · Ego4D subset (unseen environments) 3D localization | F1: 10.90 -> 12.86 | Improved |
Closest papers by meaning
- Learning Action Maps of Large Environments via First-Person Vision (CVPR 2016 · similarity 0.90)
- Egocentric Action Recognition by Automatic Relation Modeling (TPAMI 2022 · similarity 0.90)
- Ego-Only: Egocentric Action Detection without Exocentric Transferring (ICCV 2023 · similarity 0.89)
- EgoEnv: Human-centric environment representations from egocentric video (NeurIPS 2023 · similarity 0.89)
- Instance Tracking in 3D Scenes from Egocentric Videos (CVPR 2024 · similarity 0.88)
- 3D Human Pose Perception from Egocentric Stereo Videos (CVPR 2024 · similarity 0.87)
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.
