Research map / Architectures and efficiency / RNNs and normalization
Multi-Path Feedback Recurrent Neural Networks for Scene Parsing
Read the paperCheck 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 |
|---|---|---|---|
| Add multi-path feedback from the top layer to conv4, conv5, fc6, and fc7 (ablation) | VGG16 scene parser · SiftFlow | PA: 84.7 -> 86.4 | Helped |
| Add multi-path feedback from the top layer to conv4, conv5, fc6, and fc7 (ablation) | VGG16 scene parser · SiftFlow | CA: 51.5 -> 56.3 | Helped |
| Add feedback connections to fc7 only (ablation) | VGG16 scene parser · SiftFlow | PA: 84.7 -> 85.4 | Helped |
| Add feedback connections successively to conv5 and conv4 (ablation) | VGG16 scene parser · SiftFlow | PA: 85.4 -> 85.9 -> 86.2 -> 86.4 | Helped |
| Add feedback to conv3 in addition to the selected higher layers (ablation) | VGG16 scene parser · SiftFlow | PA: 86.4 -> 85.9 | Hurt |
| Use multi-step fusion across recurrent outputs rather than the final time-step feature alone (ablation) | MPF-RNN · SiftFlow, T=3 | PA: 85.2 -> 86.9 | Helped |
| Use multi-step fusion across recurrent outputs rather than the final time-step feature alone (ablation) | MPF-RNN · SiftFlow, T=3 | CA: 55.1 -> 56.5 | Helped |
| Increase recurrent steps from T=3 to T=4 (ablation) | MPF-RNN · SiftFlow | PA: 86.9 -> 86.8 | Hurt |
| Increase recurrent steps from T=3 to T=4 (ablation) | MPF-RNN · SiftFlow | CA: 56.5 -> 55.9 | Hurt |
| Use MPF-RNN instead of ResNet101-Baseline | ResNet101 scene parser · ADE20K val | PA: 73.71 -> 76.49 | Helped |
| Use MPF-RNN instead of ResNet101-Baseline | ResNet101 scene parser · ADE20K val | mIOU: 32.65 -> 34.63 | Helped |
| Use MPF-RNN instead of the strongest listed competing method | VGG16 scene parser · SiftFlow | PA: 86.8 -> 86.9 | Helped |
Closest papers by meaning
- Deep Structured Scene Parsing by Learning With Image Descriptions (CVPR 2016 · similarity 0.88)
- Learning Adaptive Receptive Fields for Deep Image Parsing Network (CVPR 2017 · similarity 0.86)
- Episodic CAMN: Contextual Attention-Based Memory Networks With Iterative Feedback for Scene Labeling (CVPR 2017 · similarity 0.86)
- DAG-Recurrent Neural Networks For Scene Labeling (CVPR 2016 · similarity 0.86)
- Making Convolutional Networks Recurrent for Visual Sequence Learning (CVPR 2018 · similarity 0.85)
- Recursive Recurrent Nets With Attention Modeling for OCR in the Wild (CVPR 2016 · similarity 0.85)
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.