Research map / Video / Action recognition
Scenes-Objects-Actions: A Multi-Task, Multi-Label Video Dataset
Jamie Ray, Heng Wang, Du Tran, Yufei Wang, Matt Feiszli, Lorenzo Torresani, Manohar Paluri
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 3D model (Res3D) instead of 2D (Res2D) | RGB video classifier · SOA | mAP (%): 23.0 -> 27.3 | Improved |
| use Res3D instead of Res2D for Actions | RGB video classifier · SOA Actions | mAP (%): 26.8 -> 33.6 | Improved |
| switch optical flow input from Res3D to Res2D | optical flow classifier · SOA Actions | mAP (%): 32.1 -> 21.5 | Hurt |
| add optical flow via late fusion with RGB | Res3D · SOA | mAP (%): 27.3 -> 30.9 | Improved |
| use optical flow instead of RGB for Scenes | Res3D · SOA Scenes | mAP (%): 48.0 -> 39.4 | Hurt |
| Pre-train on SOA rather than training from scratch (ablation) | Res3D RGB · UCF101 | accuracy (%): 67.6 -> 84.7 | Improved |
| Pre-train on SOA rather than training from scratch (ablation) | Res3D RGB · HMDB51 | accuracy (%): 33.1 -> 57.2 | Improved |
| Pre-train on SOA rather than training from scratch (ablation) | Res3D RGB · Charades | mAP (%): 9.8 -> 15.3 | Improved |
| Pre-train on SOA rather than training from scratch (ablation) | Res3D RGB · Kinetics | accuracy (%): 63.9 -> 63.9 | No clear effect |
| Pre-train on Kinetics instead of SOA | Res3D RGB · UCF101 | accuracy (%): 84.7 -> 92.6 | Improved |
| Pre-train on SOA instead of Sports-1M | Res3D RGB · Charades | mAP (%): 13.7 -> 15.3 | Improved |
| add Object features to Scene features for Scene task | Res3D RGB SVM · SOA Scenes | mAP (%): 49.7 -> 53.9 | Improved |
Closest papers by meaning
- AVA: A Video Dataset of Spatio-Temporally Localized Atomic Visual Actions (CVPR 2018 · similarity 0.91)
- Action Genome: Actions As Compositions of Spatio-Temporal Scene Graphs (CVPR 2020 · similarity 0.89)
- About Time: Advances, Challenges, and Outlooks of Action Understanding (IJCV 2025 · similarity 0.89)
- UCF-STAR: A Large Scale Still Image Dataset for Understanding Human Actions (AAAI 2020 · similarity 0.89)
- Large-Scale Weakly-Supervised Pre-Training for Video Action Recognition (CVPR 2019 · similarity 0.89)
- Action Detail Matters: Refining Video Recognition with Local Action Queries (CVPR 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.
