Research map / Video / Action recognition
AMTnet: Action-Micro-Tube Regression by End-To-End Trainable Deep Architecture
Suman Saha, Gurkirt Singh, Fabio Cuzzolin
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 |
|---|---|---|---|
| lower positive IoU threshold for 3D proposal sampling (0.5 instead of 0.7) (ablation) | AMTnet · UCF-101 | video-mAP @0.2: 54.83 -> 60.06 | Improved |
| add scheme-32 pairs to scheme-11 training data (ablation) | AMTnet · J-HMDB-21 frame-mAP | frame-mAP: 43.6 -> 45.0 | Improved |
| add scheme-32 pairs to scheme-11 training data (ablation) | AMTnet · J-HMDB-21 | video-mAP @0.5: 54.27 -> 55.31 | Improved |
| add scheme-32 pairs to scheme-11 training data (ablation) | AMTnet · J-HMDB-21 | video-mAP @0.4: 56.81 -> 56.79 | No clear effect |
| use sum fusion instead of mean fusion of feature maps (ablation) | AMTnet · J-HMDB-21 | video-mAP @0.5: 54.51 -> 55.31 | Improved |
| use top 300 instead of 1000 proposals at test time (ablation) | AMTnet · J-HMDB-21 | video-mAP @0.5: 55.31 -> 55.26 | No clear effect |
| train with consecutive-frame pairs instead of identical-frame pairs (ablation) | AMTnet · J-HMDB-21 split 1 | video-mAP @0.5: 48.9 -> 52.7 | Improved |
| train with mixed consecutive and gapped pairs instead of consecutive only (ablation) | AMTnet · J-HMDB-21 split 1 | video-mAP @0.5: 52.7 -> 57.1 | Improved |
| lower mid-layer loss weights to 0.5/0.5 instead of 1.0/1.0 (ablation) | AMTnet · J-HMDB-21 split 1 | video-mAP @0.5: 52.78 -> 53.81 | Improved |
| very low mid-layer loss weights 0.1/0.05 instead of 1.0/1.0 (ablation) | AMTnet · J-HMDB-21 split 1 | video-mAP @0.5: 52.78 -> 50.33 | Hurt |
| use 2PDP tube generation instead of micro-tube linking | AMTnet · UCF-101 | video-mAP @0.2: 61.36 -> 63.06 | Improved |
| use cross-validated alphas | AMTnet ML · UCF-101 | video-mAP @0.2: 60.06 -> 61.36 | Improved |
Later papers that compared with this method
1 later accepted paper reports AMTNet as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Change3D: Revisiting Change Detection and Captioning from A Video Modeling Perspective (CVPR 2025 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Tube Convolutional Neural Network (T-CNN) for Action Detection in Videos (ICCV 2017 · similarity 0.90)
- Recurrent Tubelet Proposal and Recognition Networks for Action Detection (ECCV 2018 · similarity 0.88)
- MiCT: Mixed 3D/2D Convolutional Tube for Human Action Recognition (CVPR 2018 · similarity 0.88)
- VideoCapsuleNet: A Simplified Network for Action Detection (NeurIPS 2018 · similarity 0.87)
- Finding Action Tubes with a Sparse-to-Dense Framework (AAAI 2020 · similarity 0.86)
- Online Human Action Detection Using Joint Classification-Regression Recurrent Neural Networks (ECCV 2016 · similarity 0.86)
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.
