Research map / Video / Action recognition
Efficient Action Recognition via Dynamic Knowledge Propagation
Hanul Kim, Mihir Jain, Jun-Tae Lee, Sungrack Yun, Fatih Porikli
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 static averaging with dynamic knowledge propagation (ablation) | student-teacher ensemble · ActivityNet 1.3 | mAP (%): 79.0 -> 81.2 | Improved |
| replace weighted-average fusion with dynamic knowledge propagation (ablation) | student-teacher ensemble · ActivityNet 1.3 | mAP (%): 80.2 -> 81.2 | Improved |
| add knowledge distillation to student (ablation) | Student · ActivityNet 1.3 | mAP (%): 77.6 -> 78.4 | Improved |
| add teacher on 5 frames at inference via dynamic propagation vs student with KD alone | Student with KD · ActivityNet 1.3 | mAP (%): 78.4 -> 81.2 | Improved |
| increase teacher frames from 5 to 20 | Teacher · ActivityNet 1.3 | mAP (%): 80.1 -> 81.7 | Improved |
| add temporal feature fusion in both student and teacher (ablation) | Dynamic-STE · ActivityNet 1.3 | mAP (%): 79.3 -> 81.2 | Improved |
| add temporal feature fusion in student only (ablation) | Dynamic-STE · ActivityNet 1.3 | mAP (%): 79.3 -> 79.9 | Improved |
| add temporal feature fusion in teacher only (ablation) | Dynamic-STE · ActivityNet 1.3 | mAP (%): 79.3 -> 80.7 | Improved |
| add frame classification loss (ablation) | Dynamic-STE · ActivityNet 1.3 | mAP (%): 80.8 -> 81.0 | Improved |
| add cosine similarity loss (ablation) | Dynamic-STE · ActivityNet 1.3 | mAP (%): 80.8 -> 80.9 | Improved |
| add both frame and cosine losses (ablation) | Dynamic-STE · ActivityNet 1.3 | mAP (%): 80.8 -> 81.2 | Improved |
| Dynamic-STE with EfficientNet vs AR-Net | efficient action recognition · ActivityNet 1.3 | mAP (%): 79.7 -> 81.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- AR-Net — AR-Net: Adaptive Frame Resolution for Efficient Action Recognition (ECCV 2020) · better on 7 of 7 shared results
- LiteEval — LiteEval: A Coarse-to-Fine Framework for Resource Efficient Video Recognition (NeurIPS 2019) · better on 2 of 2 shared results
- SCSampler — SCSampler: Sampling Salient Clips From Video for Efficient Action Recognition (ICCV 2019) · better on 2 of 2 shared results
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
- Knowledge Integration Networks for Action Recognition (AAAI 2020 · similarity 0.89)
- End-To-End Learning of Action Detection From Frame Glimpses in Videos (CVPR 2016 · similarity 0.88)
- Modular Action Concept Grounding in Semantic Video Prediction (CVPR 2022 · similarity 0.88)
- Temporal Dynamic Graph LSTM for Action-Driven Video Object Detection (ICCV 2017 · similarity 0.88)
- FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation (CVPR 2024 · similarity 0.87)
- ActionAtlas: A VideoQA Benchmark for Domain-specialized Action Recognition (NeurIPS 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.
