Research map / Video / Action recognition
EAC-Net: Efficient and Accurate Convolutional Network for Video Recognition
Bowei Jin, Zhuo Xu
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 one MGTE block at stage 3 (Res3_1) (ablation) | C2D ResNet-50 · Kinetics-400 val | top-1: 71.0 -> 72.6 | Improved |
| add one MGTE block at stage 4 instead of 3 (ablation) | C2D ResNet-50 · Kinetics-400 val | top-1: 72.6 -> 73.6 | Improved |
| add one MGTE block at stage 5 (ablation) | C2D ResNet-50 · Kinetics-400 val | top-1: 73.6 -> 73.7 | Improved |
| MGTE at both stage 4 and 5 (ablation) | C2D ResNet-50 · Kinetics-400 val | top-1: 73.7 -> 74.3 | Improved |
| also add MGTE to stage 3 (ablation) | EAC-Net w/o ATE · Kinetics-400 val | top-1: 74.3 -> 73.2 | Hurt |
| add temporal pooling branch only (ablation) | C2D · Kinetics-400 val | top-1 and FLOPs: 71.0 -> 72.0; 172.7G -> 128.2G | Improved |
| add temporal relation branch to pooling branch (ablation) | MGTE block · Kinetics-400 val | top-1: 72.0 -> 74.3 | Improved |
| add gate driven by temporal gradient (ablation) | MGTE temporal relation branch · Kinetics-400 val | top-1: 73.0 -> 73.6 | Improved |
| add global context to gate control (ablation) | MGTE gate · Kinetics-400 val | top-1: 73.6 -> 74.3 | Improved |
| replace global average pooling with ATE block (ablation) | C2D · Kinetics-400 val | top-1: 71.0 -> 72.5 | Improved |
| add ATE block (ablation) | EAC-Net MGTE stage 4+5 · Kinetics-400 val | top-1: 74.3 -> 74.8 | Improved |
| increase input from 32 to 48 frames | EAC-Net BNInception · Kinetics-400 val | top-1: 73.5 -> 74.2 | 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.
- StNet — StNet: Local and Global Spatial-Temporal Modeling for Action Recognition (AAAI 2019) · 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
- More Is Less: Learning Efficient Video Representations by Big-Little Network and Depthwise Temporal Aggregation (NeurIPS 2019 · similarity 0.85)
- TDN: Temporal Difference Networks for Efficient Action Recognition (CVPR 2021 · similarity 0.85)
- Temporal Segment Networks: Towards Good Practices for Deep Action Recognition (ECCV 2016 · similarity 0.85)
- EAN: Event Adaptive Network for Enhanced Action Recognition (IJCV 2022 · similarity 0.84)
- TEINet: Towards an Efficient Architecture for Video Recognition (AAAI 2020 · similarity 0.84)
- MEET: Towards Memory-Efficient Temporal Sparse Deep Neural Networks (CVPR 2025 · similarity 0.83)
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.
