Research map / Video / Action recognition
Memory-augmented Dense Predictive Coding for Video Representation Learning
Tengda Han, Weidi Xie, Andrew Zisserman
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 |
|---|---|---|---|
| add compressive memory (size 1024) to DPC (ablation) | DPC (R18, RGB) · UCF101 finetune | top-1 accuracy: 63.6 -> 68.2 | Improved |
| self-supervised MemDPC pretraining instead of random init (RGB) (ablation) | R18 RGB · UCF101 finetune | top-1 accuracy: 61.8 -> 68.2 | Improved |
| self-supervised MemDPC pretraining instead of random init (Flow) (ablation) | R18 Flow · UCF101 finetune | top-1 accuracy: 74.6 -> 81.9 | Improved |
| increase memory size from 512 to 1024 (ablation) | MemDPC R18 RGB · UCF101 finetune | top-1 accuracy: 65.3 -> 68.2 | Improved |
| increase memory size from 1024 to 2048 (ablation) | MemDPC R18 RGB · UCF101 finetune | top-1 accuracy: 68.2 -> 68.0 | No clear effect |
| add bi-directional aggregation (RGB) (ablation) | MemDPC R18 RGB · UCF101 finetune | top-1 accuracy: 68.2 -> 69.2 | Improved |
| add bi-directional aggregation (Flow) (ablation) | MemDPC R18 Flow · UCF101 finetune | top-1 accuracy: 81.9 -> 82.3 | Improved |
| add bi-directional aggregation (RGB+Flow) (ablation) | MemDPC R18x2 · UCF101 finetune | top-1 accuracy: 84.0 -> 84.3 | Improved |
| fuse RGB and Flow streams instead of Flow only (ablation) | MemDPC · UCF101 finetune | top-1 accuracy: 81.9 -> 84.0 | Improved |
| use non-linear probe instead of linear probe on frozen features (UCF101) | MemDPC two-stream K400 · UCF101 | top-1 accuracy: 54.1 -> 58.5 | Improved |
| use non-linear probe instead of linear probe on frozen features (HMDB51) | MemDPC two-stream K400 · HMDB51 | top-1 accuracy: 30.5 -> 33.6 | Improved |
| add two-stream fusion of RGB and Flow predictions | MemDPC K400 finetune · UCF101 | top-1 accuracy: 78.1 -> 86.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SpeedNet — SpeedNet: Learning the Speediness in Videos (CVPR 2020) · better on 1 of 1 shared result
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
- Self-Supervised Learning of Compressed Video Representations (ICLR 2021 · similarity 0.92)
- Self-Supervised Video Representation Learning by Context and Motion Decoupling (CVPR 2021 · similarity 0.90)
- Self-supervised Co-Training for Video Representation Learning (NeurIPS 2020 · similarity 0.88)
- Compressed Video Action Recognition (CVPR 2018 · similarity 0.88)
- Action Localization through Continual Predictive Learning (ECCV 2020 · similarity 0.88)
- Visual Representation Learning with Stochastic Frame Prediction (ICML 2024 · 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.
