Research map / Video / Video-language understanding
ECO: Efficient Convolutional Network for Online Video Understanding
Mohammadreza Zolfaghari, Kamaljeet Singh, Thomas Brox
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 |
|---|---|---|---|
| increase the number of sampled frames from 4 to 12 (ablation) | ECO · UCF101 | accuracy %: 90.3 -> 92.4 | Improved |
| increase the number of sampled frames from 12 to 24 (ablation) | ECO · UCF101 | accuracy %: 92.4 -> 93.6 | Improved |
| increase the number of sampled frames from 4 to 24 (ablation) | ECO · HMDB51 | accuracy %: 61.7 -> 68.4 | Improved |
| increase the number of sampled frames from 4 to 24 (ablation) | ECO · UCF101 | inference speed (VPS): 163.4 -> 28.2 | Hurt |
| add the parallel 2D stream (ECO instead of ECO Lite) at 4 frames (ablation) | ECO Lite · UCF101 | accuracy %: 87.4 -> 90.3 | Improved |
| add the parallel 2D stream (ECO instead of ECO Lite) at 4 frames (ablation) | ECO Lite · UCF101 | inference speed (VPS): 237.3 -> 163.4 | Hurt |
| add the parallel 2D stream (ECO instead of ECO Lite) at 4 frames (ablation) | ECO Lite · HMDB51 | accuracy %: 58.1 -> 61.7 | Improved |
| add the parallel 2D stream on Something-Something (ablation) | ECO Lite · Something-Something | accuracy % | No clear effect |
| use more sampled frames for simple short-term actions (ablation) | ECO · UCF101/HMDB51 simple actions | per-class accuracy | Hurt |
| add optical flow to RGB in the ensemble (ablation) | ECOEnLite · Something-Something val | accuracy %: 46.4 -> 49.5 | Improved |
| add optical flow to RGB in the ensemble (ablation) | ECOEnLite · Something-Something test | accuracy %: 42.3 -> 43.9 | Improved |
| replace M-TRN with ECOEnLite | M-TRN · Something-Something val | accuracy %: 34.44 -> 46.4 | Improved |
Closest papers by meaning
- Temporal Query Networks for Fine-Grained Video Understanding (CVPR 2021 · similarity 0.87)
- More Is Less: Learning Efficient Video Representations by Big-Little Network and Depthwise Temporal Aggregation (NeurIPS 2019 · similarity 0.87)
- Beyond Short Clips: End-to-End Video-Level Learning With Collaborative Memories (CVPR 2021 · similarity 0.87)
- Convolutional Two-Stream Network Fusion for Video Action Recognition (CVPR 2016 · similarity 0.87)
- No Frame Left Behind: Full Video Action Recognition (CVPR 2021 · similarity 0.87)
- Learning End-to-end Video Classification with Rank-Pooling (ICML 2016 · 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.
