Awesome

Efficient-Segmentation-Networks

This project aims at providing an easy-to-use, modifiable reference implementation for real-time semantic segmentation models using PyTorch.

Table of Contents:

• <a href='#Requirements'>Requirements</a>
• <a href='#Models'>Models</a>
• <a href='#Dataset-Setting'>Dataset Setting</a>
• <a href='#Usage'>Usage</a>
• <a href='#Contact'>Contact</a>

Requirements

PyTorch and Torchvision needs to be installed before running the scripts, PyTorch v1.1 or later is supported.

pip3 install -r requirements.txt

Models

The project supports these semantic segmentation models as follows:

• (SQNet) Speeding up Semantic Segmentation for Autonomous Driving [Paper]
• (LinkNet) Exploiting Encoder Representations for Efficient Semantic Segmentation [Paper]
• (SegNet) A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation [Paper]
• (UNet) Convolutional Networks for Biomedical Image Segmentation [Paper]
• (ENet) A Deep Neural Network Architecture for Real-Time Semantic Segmentation [Paper]
• (ERFNet) Efficient ConvNet for Real-time Semantic Segmentation [Paper]
• (EDANet) Efficient Dense Modules of Asymmetric Convolution for Real-Time Segmentation [Paper]
• (ESPNet) Efficient Spatial Pyramid of Dilated Convolutions for Semantic Segmentation [Paper]
• (ESPNetv2) A Light-weight, Power Efficient, and General Purpose ConvNet [Paper]
• (LEDNet) A Lightweight Encoder-Decoder Network for Real-Time Semantic Segmentation [Paper]
• (FSSNet) Fast Semantic Segmentation for Scene Perception [Paper]
• (ESNet) An Efficient Symmetric Network for Real-time Semantic Segmentation [Paper]
• (CGNet) A Light-weight Context Guided Network for Semantic Segmentation [Paper]
• (Fast-SCNN) Fast Semantic Segmentation Network [Paper]
• (DABNet) Depth-wise Asymmetric Bottleneck for Real-time Semantic Segmentation [Paper]
• (ContextNet) Exploring Context and Detail for Semantic Segmentation in Real-time [Paper]
• (FPENet) Feature Pyramid Encoding Network for Real-time Semantic Segmentation [Paper]
• ...

Losses

The project supports these loss functions:

1. Weighted Cross Entropy
2. Weighted Cross Entropy with Label Smooth
3. Focal Loss
4. Ohem Cross Entropy
5. LovaszSoftmax
6. SegLoss-List
7. ...

Optimizers

The project supports these optimizers:

1. SGD
2. Adam
3. AdamW
4. RAdam
5. RAdam + Lookahead
6. ...

Activations

1. ReLu
2. PReLU
3. ReLU6
4. Swish
5. Mish : A Self Regularized Non-Monotonic Neural Activation Function
6. ...

Learning Rate Scheduler

The project supports these LR_Schedulers:

1. Poly decay
2. Warmup Poly
3. ...

Normalization methods

1. In-Place Activated BatchNorm
2. Switchable Normalization
3. Weight Standardization
4. ...

Enhancing Semantic Feature Learning Method

1. Attention Family
2. NAS Family
3. ...

Some useful Tools

1. pytorch-OpCounter
2. flops-counter.pytorch
3. Netron : Visualizer for neural network models, On line URL: Netron
4. Falshtorch: Visualization toolkit for neural networks in PyTorch !
5. Bag of Tricks for Image Classification with Convolutional Neural Networks
6. ...

Dataset-Setting

This project has been tailored to suit the Cityscapes and CamVid datasets. The folds of your dataset need satisfy the following structures:

|-- dataset
|  |-- camvid
|  |  |-- train
|  |  |-- trainannot
|  |  |-- val
|  |  |-- valannot
|  |  |-- test
|  |  |-- testannot
|  |  |-- ...
|  |-- cityscapes
|  |  |-- leftImg8bit
|  |  |  |-- train
|  |  |  |-- val
|  |  |  |-- test
|  |  |-- gtFine
|  |  |  |-- train
|  |  |  |-- val
|  |  |  |-- test
|  |  |-- ...

• You can download cityscapes dataset from here. Note: please download leftImg8bit_trainvaltest.zip(11GB) and gtFine_trainvaltest(241MB).
• You can download camvid dataset from here.
• The Cityscapes dataset scripts for inspection, preparation, and evaluation can download from here.

Usage

Clone this Repo

git clone https://github.com/xiaoyufenfei/Efficient-Segmentation-Networks
cd Efficient-Segmentation-Networks

Currently, the code supports Python 3

Torch dependencies:

• PyTorch (>=1.1.0)
• torchvision(>=0.3.0)

Data dependencies:

• Cityscapes + scripts

Download Cityscapes and run the script createTrainIdLabelImgs.py to create annotations based on the training labels. Make sure that the folder is named cityscapes

Training

• For Cityscapes / CamVid

1. training on train set

python train.py  --help

2. training on train+val set

python train.py --help

Testing

• For Cityscapes / CamVid

python test.py --help

Predicting

• For Cityscapes

python predict.py --help

Evaluating

• For Cityscapes

cd tools
python trainID2labelID.py 

Contact

If you think this work useful, please give me a star! And if you find any errors or have any suggestions, please contact me.

GitHub: xiaoyufenfei Email: wangy314159@163.com

Refer to this Rep

You are encouraged to cite the following papers if this work helps your research.

@misc{Efficient-Segmentation-Networks,
  author = {Yu Wang},
  title = {Efficient-Segmentation-Networks Pytorch Implementation},
  year = {2019},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/xiaoyufenfei/Efficient-Segmentation-Networks}},
  commit = {master}
}

License

This project is released under the MIT License. See LICENSE for additional details.