Research map / Detection and segmentation / CNN features
DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs
Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin P. Murphy, Alan Yuille
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 dense CRF post-processing (ablation) | DeepLab VGG-16 variants · PASCAL VOC 2012 val | mIOU: 64.38 -> 67.64 | Improved |
| use LargeFOV (3x3 kernel, r=12) instead of 7x7 kernel (ablation) | DeepLab VGG-16 · PASCAL VOC 2012 val | mIOU after CRF: 67.64 -> 67.64 | No clear effect |
| use poly instead of step learning rate policy (ablation) | DeepLab-LargeFOV · PASCAL VOC 2012 val | mIOU: 62.25 -> 63.42 | Improved |
| train longer (6K -> 10K iterations) with poly (ablation) | DeepLab-LargeFOV · PASCAL VOC 2012 val | mIOU: 63.42 -> 64.90 | Improved |
| reduce batch size from 30 to 10 (ablation) | DeepLab-LargeFOV · PASCAL VOC 2012 val | mIOU: 64.90 -> 64.71 | No clear effect |
| add ASPP-S (rates 2,4,8,12) over LargeFOV (ablation) | DeepLab VGG-16 · PASCAL VOC 2012 val | mIOU: 65.76 -> 66.98 before CRF; 69.84 -> 69.73 after CRF | Mixed |
| add ASPP-L (rates 6,12,18,24) over LargeFOV (ablation) | DeepLab VGG-16 · PASCAL VOC 2012 val | mIOU: 65.76 -> 68.96 before CRF; 69.84 -> 71.57 after CRF | Improved |
| replace VGG-16 with ResNet-101 (ablation) | DeepLab · PASCAL VOC 2012 val | mIOU before CRF: 65.76 -> 68.72 | Improved |
| add multi-scale inputs with max fusion (ablation) | DeepLab ResNet-101 · PASCAL VOC 2012 val | mIOU: 68.72 -> 71.27 | Improved |
| add MS-COCO pretraining (ablation) | DeepLab ResNet-101 · PASCAL VOC 2012 val | mIOU: 71.27 -> 73.28 | Improved |
| add random-scale data augmentation (ablation) | DeepLab ResNet-101 · PASCAL VOC 2012 val | mIOU: 73.28 -> 74.87 | Improved |
| add LargeFOV layer (ablation) | DeepLab ResNet-101 · PASCAL VOC 2012 val | mIOU: 74.87 -> 75.54 | Improved |
Closest papers by meaning
- Fully Convolutional Networks for Semantic Segmentation (TPAMI 2016 · similarity 0.92)
- DenseASPP for Semantic Segmentation in Street Scenes (CVPR 2018 · similarity 0.91)
- Revisiting Dilated Convolution: A Simple Approach for Weakly- and Semi-Supervised Semantic Segmentation (CVPR 2018 · similarity 0.91)
- Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation (ECCV 2018 · similarity 0.91)
- Segmentation-Aware Convolutional Networks Using Local Attention Masks (ICCV 2017 · similarity 0.91)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 2024 · similarity 0.90)
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.
