Research map / Video / Action recognition
Multi-Moments in Time: Learning and Interpreting Models for Multi-Action Video Understanding
Mathew Monfort, Bowen Pan, Kandan Ramakrishnan, Alex Andonian, Barry McNamara, Alex Lascelles, Quanfu Fan, Dan Gutfreund, Rogério Feris, Aude Oliva
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 |
|---|---|---|---|
| use class-weighted per-positive wLSEP instead of LSEP (ablation) | I3D ResNet-50 · M-MiT validation | top-1 accuracy: 52.7 -> 58.5 | Improved |
| use wLSEP instead of BCE (ablation) | I3D ResNet-50 · M-MiT validation | micro mAP: 55.4 -> 61.7 | Improved |
| use wLSEP instead of BCE (ablation) | TSM · M-MiT validation | macro mAP: 37.8 -> 35.4 | Hurt |
| use wLSEP instead of BCE (ablation) | I3D ResNet-50 · M-MiT validation | macro mAP: 37.8 -> 39.4 | Improved |
| use wLSEP instead of WARP (ablation) | I3D ResNet-50 · AVA | top-1 accuracy: 47.6 -> 71.2 | Improved |
| use wLSEP instead of BCE (ablation) | I3D ResNet-50 · MultiThumos | macro mAP: 58.9 -> 68.0 | Improved |
| fuse SoundNet audio features with I3D (ablation) | I3D wLSEP · M-MiT validation | top-1 accuracy: 58.5 -> 59.3 | No clear effect |
| add extra M-MiT labels to evaluation (ablation) | MiT-trained I3D · MiT shared-class subset | top-1 accuracy: 23.0 -> 40.0 | Improved |
| train on multi-label M-MiT instead of MiT | I3D ResNet-50 · M-MiT-S | top-1 accuracy: 40.0 -> 45.0 | Improved |
| pretrain on M-MiT instead of MiT, fine-tune on MultiTHUMOS | I3D ResNet-50 · MultiTHUMOS | top-1 accuracy: 0.849 -> 0.873 | Improved |
| pretrain on M-MiT instead of MiT, fine-tune on HMDB | I3D ResNet-50 · HMDB | top-1 accuracy: 0.756 -> 0.740 | Hurt |
| pretrain on M-MiT instead of MiT, fine-tune on UCF-101 | I3D ResNet-50 · UCF-101 | top-1 accuracy: 0.908 -> 0.892 | Hurt |
Closest papers by meaning
- Every Moment Counts: Dense Detailed Labeling of Actions in Complex Videos (IJCV 2017 · similarity 0.91)
- PointTAD: Multi-Label Temporal Action Detection with Learnable Query Points (NeurIPS 2022 · similarity 0.90)
- An Effective-Efficient Approach for Dense Multi-Label Action Detection (IJCV 2026 · similarity 0.90)
- Weakly-Supervised Temporal Action Detection for Fine-Grained Videos with Hierarchical Atomic Actions (ECCV 2022 · similarity 0.90)
- About Time: Advances, Challenges, and Outlooks of Action Understanding (IJCV 2025 · similarity 0.89)
- SeFAR: Semi-supervised Fine-grained Action Recognition with Temporal Perturbation and Learning Stabilization (AAAI 2025 · 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.
