Research map / Video / Action recognition
Deep Feature Flow for Video Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| propagate key-frame features with end-to-end jointly trained flow and recognition networks | DFF versus Frame · Cityscapes validation | runtime (fps): 1.52 -> 5.60 | Helped |
| propagate key-frame features with end-to-end jointly trained flow and recognition networks | DFF versus Frame · Cityscapes validation | mIoU (%): 71.1 -> 69.2 | Hurt |
| propagate key-frame features with end-to-end jointly trained flow and recognition networks | DFF versus Frame · ImageNet VID validation | runtime (fps): 4.05 -> 20.25 | Helped |
| propagate key-frame features with end-to-end jointly trained flow and recognition networks | DFF versus Frame · ImageNet VID validation | mAP (%): 73.9 -> 73.1 | Hurt |
| use trained feature flow instead of off-the-shelf SIFT-Flow | DFF versus SFF-fast and SFF-slow · Cityscapes and ImageNet VID | recognition accuracy: Cityscapes mIoU: 67.3 -> 69.2; ImageNet VID mAP: 69.7 -> 73.1 | Helped |
| jointly train the image recognition network rather than fixing it (ablation) | DFF · Cityscapes validation | mIoU (%): 68.8 -> 69.2 | Helped |
| jointly train the image recognition network rather than fixing it (ablation) | DFF · ImageNet VID validation | mAP (%): 72.3 -> 73.1 | Helped |
| jointly train the flow network rather than fixing it (ablation) | DFF · Cityscapes validation | mIoU (%): 67.0 -> 69.2 | Helped |
| jointly train the flow network rather than fixing it (ablation) | DFF · ImageNet VID validation | mAP (%): 68.8 -> 73.1 | Helped |
| jointly train the recognition and flow networks instead of training them separately (ablation) | DFF · Cityscapes validation | mIoU (%): 66.9 -> 69.2 | Helped |
| jointly train the recognition and flow networks instead of training them separately (ablation) | DFF · ImageNet VID validation | mAP (%): 67.4 -> 73.1 | Helped |
| add the scale function to modulate propagated features (ablation) | DFF · Cityscapes and ImageNet VID | recognition accuracy | Helped |
Closest papers by meaning
- Optical Flow Guided Feature: A Fast and Robust Motion Representation for Video Action Recognition (CVPR 2018 · similarity 0.89)
- Representation Flow for Action Recognition (CVPR 2019 · similarity 0.88)
- Deep Residual Learning for Image Recognition (CVPR 2016 · similarity 0.87)
- Real-Time Action Recognition With Enhanced Motion Vector CNNs (CVPR 2016 · similarity 0.87)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.86)
- FlowFeat: Pixel-Dense Embedding of Motion Profiles (NeurIPS 2025 · similarity 0.86)
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.