linbo0518/BLSeg

PyTorch's Semantic Segmentation Toolbox

BLSeg (BaseLine Segmentation)

PyTorch's Semantic Segmentation Toolbox

• BLSeg (BaseLine Segmentation)
  • Requirement
  • Quick Start
  • Documentation
  • Supported Module
  • Analysis
  • Visualization
  • Changelog
  • References

Requirement

• Python 3
• PyTorch >= 1.0.0

Quick Start

Execute the following command in your terminal

pip install --upgrade git+https://github.com/linbo0518/BLSeg.git

Documentation

For more information, please see Documentation

Supported Module

• Backbone
  • VGG16
  • VGG19
  • MobileNet v1 (1.0)
  • MobileNet v2 (1.0)
  • ResNet 34
  • ResNet 50 (Modified according to Bag of Tricks)
  • SE ResNet 34
  • SE ResNet 50 (Modified according to Bag of Tricks)
  • Modified Aligned Xception
• Model
  • FCN
  • U-Net
  • PSPNet
  • DeepLab v3+
  • GCN (Large Kernel Matters)
• Loss
  • BCEWithLogitsLossWithOHEM
  • CrossEntropyLossWithOHEM
  • DiceLoss (only for binary classification)
  • SoftCrossEntropyLossWithOHEM
• Metric
  • Loss Meter
  • Pixel Accuracy
  • Mean IoU
• Others
  • Xavier/MSRA initialization (support zero gamma in last BatchNorm)
  • Pre-trained weight
  • Mixed precision training
  • Online Hard Example Mining
  • Precise BatchNorm
  • Freeze/train Backbone model
  • Freeze/train BatchNorm layers

Each segmentation model can combine with any backbone without any modifications.

Backbone \ Model	FCN	U-Net	PSPNet	DeepLab v3+	GCN
VGG16	√	√	√	√	√
VGG19	√	√	√	√	√
MobileNet v1	√	√	√	√	√
MobileNet v2	√	√	√	√	√
ResNet34	√	√	√	√	√
ResNet50	√	√	√	√	√
SE ResNet34	√	√	√	√	√
SE ResNet50	√	√	√	√	√
Modified Aligned Xception	√	√	√	√	√

Model pre-trained on augmented PASCAL VOC2012 dataset with 10582 images for training and 1449 images for validation.

You can download pre-trained parameters at Google Drive

Analysis

• Parameters

Backbone \ Model	FCN	U-Net	PSPNet	DeepLab v3+	GCN
VGG16	134.82M	25.26M	19.71M	20.15M	15.36M
VGG19	140.13M	30.57M	25.02M	25.46M	20.67M
MobileNet v1	226.07M	14.01M	13.71M	12.44M	4.04M
MobileNet v2	276.46M	2.68M	15.67M	13.36M	2.88M
ResNet34	141.38M	24.08M	26.28M	26.72M	21.76M
ResNet50	451.92M	66.35M	46.61M	40.37M	25.09M
SE ResNet34	141.54M	24.25M	26.44M	26.87M	21.92M
SE ResNet50	454.44M	69.03M	49.13M	42.89M	27.61M
Modified Aligned Xception	466.25M	57.46M	60.95M	54.71M	39.17M

• Multiply-accumulate operations (MACs)

Backbone \ Model	FCN	U-Net	PSPNet	DeepLab v3+	GCN
VGG16	348.11G	114.79G	121.38G	121.38G	85.42G
VGG19	390.27G	136.54G	103.21Gs	127.44G	107.17G
MobileNet v1	228.52G	37.83G	52.41G	52.41G	8.24G
MobileNet v2	240.77G	3.22G	58.00G	58.00G	5.80G
ResNet34	230.69G	34.99G	109.88G	109.88G	23.65G
ResNet50	326.75G	133.01G	178.64G	178.64G	29.44G
SE ResNet34	230.70G	35.00G	109.90G	109.90G	23.66G
SE ResNet50	326.81G	133.05G	178.72G	178.72G	29.47G
Modified Aligned Xception	359.83G	83.46G	237.01G	237.01G	44.88G

Visualization

Original Image	Ground Truth	Ours

Changelog

See Changelog

References

• Simonyan, Karen, and Andrew Zisserman. "Very deep convolutional networks for large-scale image recognition." arXiv preprint arXiv:1409.1556 (2014).
• Howard, Andrew G., et al. "Mobilenets: Efficient convolutional neural networks for mobile vision applications." arXiv preprint arXiv:1704.04861 (2017).
• Sandler, Mark, et al. "Mobilenetv2: Inverted residuals and linear bottlenecks." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018.
• He, Kaiming, et al. "Deep residual learning for image recognition." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.
• Xie, Junyuan, et al. "Bag of tricks for image classification with convolutional neural networks." arXiv preprint arXiv:1812.01187 (2018).
• Hu, Jie, Li Shen, and Gang Sun. "Squeeze-and-excitation networks." Proceedings of the IEEE conference on computer vision and pattern recognition. 2018.
• Long, Jonathan, Evan Shelhamer, and Trevor Darrell. "Fully convolutional networks for semantic segmentation." Proceedings of the IEEE conference on computer vision and pattern recognition. 2015.
• Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. "U-net: Convolutional networks for biomedical image segmentation." International Conference on Medical image computing and computer-assisted intervention. Springer, Cham, 2015.
• Zhao, Hengshuang, et al. "Pyramid scene parsing network." Proceedings of the IEEE conference on computer vision and pattern recognition. 2017.
• Chen, Liang-Chieh, et al. "Encoder-decoder with atrous separable convolution for semantic image segmentation." Proceedings of the European Conference on Computer Vision (ECCV). 2018.
• Peng, Chao, et al. "Large Kernel Matters--Improve Semantic Segmentation by Global Convolutional Network." Proceedings of the IEEE conference on computer vision and pattern recognition. 2017.

---