Research map / Video / Action recognition
LiteEval: A Coarse-to-Fine Framework for Resource Efficient Video Recognition
Zuxuan Wu, Caiming Xiong, Yu-Gang Jiang, Larry S. Davis
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 |
|---|---|---|---|
| Use adaptive coarse-to-fine feature computation instead of uniform frame evaluation | LiteEval · FCVID | GFLOPs: 195.5 -> 94.3 | Improved |
| Use adaptive coarse-to-fine feature computation instead of uniform frame evaluation | LiteEval · ActivityNet | mAP: 70.0% -> 72.7% | Improved |
| Use adaptive coarse-to-fine feature computation instead of uniform frame evaluation | LiteEval · ActivityNet | GFLOPs: 195.5 -> 95.1 | Improved |
| Replace Gumbel-Softmax gating with REINFORCE policy learning | LiteEval · FCVID | mAP: 80.0% -> 74.2% | Hurt |
| Replace Gumbel-Softmax gating with REINFORCE policy learning | LiteEval · ActivityNet | GFLOPs: 95.1 -> 269.3 | Hurt |
| Add synchronization that transfers coarse LSTM information to the fine LSTM (ablation) | LiteEval · FCVID | mAP: 65.7% -> 80.0% | Improved |
| Set gamma to 0.01 rather than 0.05 (ablation) | LiteEval · FCVID | mAP: 80.0% -> 78.8% | Hurt |
| Set gamma to 0.03 rather than 0.05 (ablation) | LiteEval · FCVID | mAP: 80.0% -> 79.7% | Hurt |
| Set gamma to 0.10 rather than 0.05 (ablation) | LiteEval · FCVID | GFLOPs: 94.3 -> 139.0 | Hurt |
| Use 64 cLSTM hidden units instead of 512 (ablation) | LiteEval · FCVID | mAP: 80.0% -> 76.9% | Hurt |
| Use 128 cLSTM hidden units instead of 512 (ablation) | LiteEval · FCVID | mAP: 80.0% -> 77.3% | Hurt |
| Use 256 cLSTM hidden units instead of 512 (ablation) | LiteEval · FCVID | mAP: 80.0% -> 78.3% | Hurt |
Later papers that compared with this method
6 later accepted papers report LiteEval as a baseline in their results tables; 5 of them report a better number on most of the results they share with it.
- Facial Expression Recognition with Adaptive Frame Rate based on Multiple Testing Correction (ICML 2023 · better on 0 of 1 shared result)
- SpotEM: Efficient Video Search for Episodic Memory (ICML 2023 · better on 1 of 1 shared result)
- Dynamic Network Quantization for Efficient Video Inference (ICCV 2021 · better on 4 of 4 shared results)
- Efficient Action Recognition via Dynamic Knowledge Propagation (ICCV 2021 · better on 2 of 2 shared results)
- SMART Frame Selection for Action Recognition (AAAI 2021 · better on 1 of 1 shared result)
- AR-Net: Adaptive Frame Resolution for Efficient Action Recognition (ECCV 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
- A Coarse-to-Fine Framework for Resource Efficient Video Recognition (IJCV 2021 · similarity 0.99)
- Principles of Visual Tokens for Efficient Video Understanding (ICCV 2025 · similarity 0.85)
- AdaFrame: Adaptive Frame Selection for Fast Video Recognition (CVPR 2019 · similarity 0.85)
- Uni-AdaFocus: Spatial-Temporal Dynamic Computation for Video Recognition (TPAMI 2024 · similarity 0.85)
- VLG: General Video Recognition with Web Textual Knowledge (IJCV 2024 · similarity 0.85)
- LookWhen? Fast Video Recognition by Learning When, Where, and What to Compute (NeurIPS 2026 · 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.
