Research map / Video / Action recognition
Action Prediction From Videos via Memorizing Hard-to-Predict Samples
Yu Kong, Shangqian Gao, Bin Sun, Yun Fu
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 memory module of hard-to-predict samples (ablation) | mem-LSTM two-stream · UCF-101, observation ratio 0.5 | accuracy: 83.16% -> 88.37% | Improved |
| add memory module of hard-to-predict samples (ablation) | mem-LSTM two-stream · UCF-101, observation ratio 1.0 | accuracy: 85.80% -> 90.49% | Improved |
| add residual connections to 2-layer LSTM (ablation) | RGB LSTM · UCF-101, observation ratio 0.5 | accuracy: 71.58% -> 72.51% | Improved |
| add backward LSTM layer (bi-directional) (ablation) | RGB LSTM (l=2) · UCF-101, observation ratio 1.0 | accuracy: 73.15% -> 75.47% | Improved |
| stack a second LSTM layer (ablation) | RGB LSTM · UCF-101, observation ratio 1.0 | accuracy: 71.37% -> 73.15% | Improved |
| add flow stream to RGB stream | RGB Bi-LSTM · UCF-101, observation ratio 0.1 | accuracy: 50.01% -> 51.02% | Improved |
| use bi-directional LSTM on flow stream (ablation) | flow LSTM · UCF-101 | accuracy | Hurt |
| use residual connection and regularization on flow stream (ablation) | flow LSTM · UCF-101 | accuracy | Hurt |
| fuse streams by average, sum or max pooling instead of concatenation (ablation) | mem-LSTM · UCF-101 | accuracy | Hurt |
| ImageNet pre-train ResNet-18 before fine-tuning (ablation) | ResNet-18 · UCF-101 | accuracy: 47.9% -> 69.1% | Improved |
| mem-LSTM vs DeepSCN | mem-LSTM · Sports-1M, observation ratio 0.1 | accuracy: 55.02% -> 57.60% | Improved |
Closest papers by meaning
- Lattice Long Short-Term Memory for Human Action Recognition (ICCV 2017 · similarity 0.89)
- Regularizing Long Short Term Memory With 3D Human-Skeleton Sequences for Action Recognition (CVPR 2016 · similarity 0.89)
- Deep Sequential Context Networks for Action Prediction (CVPR 2017 · similarity 0.88)
- Eidetic 3D LSTM: A Model for Video Prediction and Beyond (ICLR 2019 · similarity 0.87)
- Encouraging LSTMs to Anticipate Actions Very Early (ICCV 2017 · similarity 0.87)
- Adversarial Action Prediction Networks (TPAMI 2018 · 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.
