Research map / Video / Action recognition
Self-Supervised Spatiotemporal Learning via Video Clip Order Prediction
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train C3D with clip order prediction rather than random initialization | C3D · UCF101 | action recognition accuracy: 61.6 -> 65.6 | Helped |
| Train R3D with clip order prediction rather than random initialization | R3D · UCF101 | action recognition accuracy: 54.4 -> 64.9 | Helped |
| Train R(2+1)D with clip order prediction rather than random initialization | R(2+1)D · UCF101 | action recognition accuracy: 56.2 -> 72.4 | Helped |
| Train C3D with clip order prediction rather than random initialization | C3D · HMDB51 | action recognition accuracy: 23.2 -> 28.4 | Helped |
| Train R3D with clip order prediction rather than random initialization | R3D · HMDB51 | action recognition accuracy: 21.5 -> 29.5 | Helped |
| Train R(2+1)D with clip order prediction rather than random initialization | R(2+1)D · HMDB51 | action recognition accuracy: 22.0 -> 30.9 | Helped |
| Use clip order prediction training instead of random initialization | C3D · UCF101 split 1 retrieval | Top-1 retrieval accuracy: 16.0 -> 12.5 | Hurt |
| Use clip order prediction training instead of random initialization | C3D · UCF101 split 1 retrieval | Top-50 retrieval accuracy: 39.3 -> 66.9 | Helped |
| Use clip order prediction training instead of random initialization | R3D · UCF101 split 1 retrieval | Top-50 retrieval accuracy: 36.7 -> 66.5 | Helped |
| Use clip order prediction training instead of random initialization | R(2+1)D · UCF101 split 1 retrieval | Top-50 retrieval accuracy: 38.5 -> 63.9 | Helped |
| Use clip order prediction training instead of random initialization | C3D · HMDB51 split 1 retrieval | Top-50 retrieval accuracy: 37.8 -> 70.1 | Helped |
| Use clip order prediction training instead of random initialization | R3D · HMDB51 split 1 retrieval | Top-50 retrieval accuracy: 37.2 -> 68.9 | Helped |
Closest papers by meaning
- A Closer Look at Spatiotemporal Convolutions for Action Recognition (CVPR 2018 · similarity 0.88)
- Self-Supervised Spatio-Temporal Representation Learning for Videos by Predicting Motion and Appearance Statistics (CVPR 2019 · similarity 0.88)
- Spatiotemporal Fusion in 3D CNNs: A Probabilistic View (CVPR 2020 · similarity 0.87)
- Spatiotemporal Contrastive Video Representation Learning (CVPR 2021 · similarity 0.87)
- Collaborative Spatiotemporal Feature Learning for Video Action Recognition (CVPR 2019 · similarity 0.87)
- Compositional Video Understanding with Spatiotemporal Structure-based Transformers (CVPR 2024 · similarity 0.85)
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.