Research map / Video / Action recognition
Self-Supervised Video Representation Learning with Space-Time Cubic Puzzles
Read the paperCheck 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 |
|---|---|---|---|
| Pretrain with Space-Time Cubic Puzzles rather than random initialization | 3D ResNet-18 · UCF101 | top-1 accuracy: 42.4 -> 65.8 | Helped |
| Pretrain with Space-Time Cubic Puzzles rather than random initialization | 3D ResNet-18 · HMDB51 | top-1 accuracy: 17.1 -> 33.7 | Helped |
| Use Space-Time Cubic Puzzles instead of Kinetics 1/8 pretraining | 3D ResNet-18 · UCF101 | top-1 accuracy: 64.2 -> 65.8 | Helped |
| Use Space-Time Cubic Puzzles instead of Kinetics 1/8 pretraining | 3D ResNet-18 · HMDB51 | top-1 accuracy: 33.2 -> 33.7 | Helped |
| Use Space-Time Cubic Puzzles instead of ImageNet-inflated pretraining | 3D ResNet-18 · UCF101 | top-1 accuracy: 60.3 -> 65.8 | Helped |
| Use Space-Time Cubic Puzzles instead of ImageNet-inflated pretraining | 3D ResNet-18 · HMDB51 | top-1 accuracy: 30.7 -> 33.7 | Helped |
| Combine spatial and temporal puzzle training in one method rather than late-fusing separate puzzle scores | 3D ResNet-18 · UCF101 | top-1 accuracy: 61.3 -> 65.8 | Helped |
| Add channel replication (ablation) | Space-Time Cubic Puzzles · UCF101 | top-1 accuracy: 58.7 -> 61.5 | Helped |
| Add random spatio-temporal jittering (ablation) | Space-Time Cubic Puzzles with channel replication · UCF101 | top-1 accuracy: 61.5 -> 63.9 | Helped |
| Add rotation with classification (ablation) | Space-Time Cubic Puzzles with channel replication and jittering · UCF101 | top-1 accuracy: 63.9 -> 65.8 | Helped |
| Use spatial and temporal puzzle contexts together rather than spatial-only puzzles | 3D ResNet-18 · UCF101 | top-1 accuracy: 58.5 -> 65.8 | Helped |
| Use spatial and temporal puzzle contexts together rather than temporal-only puzzles | 3D ResNet-18 · UCF101 | top-1 accuracy: 59.3 -> 65.8 | Helped |
Closest papers by meaning
- Self-Supervised Spatio-Temporal Representation Learning for Videos by Predicting Motion and Appearance Statistics (CVPR 2019 · similarity 0.87)
- Geometry Guided Convolutional Neural Networks for Self-Supervised Video Representation Learning (CVPR 2018 · similarity 0.85)
- Video Cloze Procedure for Self-Supervised Spatio-Temporal Learning (AAAI 2020 · similarity 0.85)
- Cross-Architecture Self-Supervised Video Representation Learning (CVPR 2022 · similarity 0.84)
- Spatial-Then-Temporal Self-Supervised Learning for Video Correspondence (CVPR 2023 · similarity 0.84)
- Chirality in Action: Time-Aware Video Representation Learning by Latent Straightening (NeurIPS 2025 · similarity 0.83)
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.