Research map / Video / Action recognition
Action Anticipation with RBF Kernelized Feature Mapping RNN
Yuge Shi, Basura Fernando, Richard Hartley
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 |
|---|---|---|---|
| replace vanilla LSTM with feature mapping RNN (ablation) | RBF-free predictor · JHMDB-21 variants (Inception V3, p=50%, max pooling) | accuracy: 66.3% -> 72.2% | Improved |
| replace linear D x D mapping with feature mapping RNN (ablation) | feature generator · JHMDB-21 variants (Inception V3, p=50%, max pooling) | accuracy: 62.7% -> 72.2% | Improved |
| add RBF kernel layer (ablation) | Feature Mapping RNN · JHMDB-21 variants (Inception V3, p=50%, max pooling) | accuracy: 72.2% -> 72.8% | Improved |
| add adversarial loss (ablation) | Feature Mapping RNN + RBF · JHMDB-21 variants (Inception V3, p=50%, max pooling) | accuracy: 72.8% -> 73.4% | Improved |
| add RBF kernel layer (ablation) | Vanilla LSTM · JHMDB-21 variants (Inception V3, p=50%, max pooling) | accuracy: 66.3% -> 67.9% | Improved |
| add adversarial loss (ablation) | Vanilla LSTM + RBF · JHMDB-21 variants (Inception V3, p=50%, max pooling) | accuracy | Hurt |
| switch features from VGG16 to Inception V3 (ablation) | Feature Mapping RNN · JHMDB-21 anticipation (20% observed) | accuracy: 63% -> 70% | Improved |
| add kernel SVM and adversarial training (ablation) | Feature Mapping RNN (VGG16) · JHMDB-21 anticipation (20% observed) | accuracy: 63% -> 67% | Improved |
| replace ELSTM with feature mapping RNN on same VGG16 features | action anticipation · JHMDB-21 anticipation (20% observed) | accuracy: 55% -> 63% | Improved |
| reduce feature stride from 128 to 4 (ablation) | Feature Mapping RNN · JHMDB-21 feature stride sweep | accuracy: 72.4% -> 74.3% | Improved |
| increase LSTM hidden state size from 2 to 64 (ablation) | Feature Mapping RNN · JHMDB-21 LSTM hidden state sweep | accuracy: 71.7% -> 73.8% | No clear effect |
| increase number of RBF kernels from 8 to 128 (ablation) | Feature Mapping RNN · JHMDB-21 RBF kernel count sweep | accuracy: 72.7% -> 73.8% | Improved |
Closest papers by meaning
- Predicting the Future: A Jointly Learnt Model for Action Anticipation (ICCV 2019 · similarity 0.88)
- Encouraging LSTMs to Anticipate Actions Very Early (ICCV 2017 · similarity 0.88)
- Early Action Prediction by Soft Regression (TPAMI 2018 · similarity 0.87)
- Rethinking Learning Approaches for Long-Term Action Anticipation (ECCV 2022 · similarity 0.86)
- Time-Conditioned Action Anticipation in One Shot (CVPR 2019 · similarity 0.86)
- Can't Make an Omelette Without Breaking Some Eggs: Plausible Action Anticipation Using Large Video-Language Models (CVPR 2024 · similarity 0.84)
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.
