Implementation of ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation using PyTorch (v.1.10)
This implementation is validated on the CamVid dataset. The pre-trained of ENet trained with CamVid is available here
| Dataset | Type | Classes | Input resolution | Batch size | Epochs | Mean IoU (%) | GFLOPS | Parameters |
|---|---|---|---|---|---|---|---|---|
| CamVid | Reproduced | 11 | 480x360 | 10 | 100 | 58.6 | 2.34 | 0.35M |
| Cityscapes | Reproduced | 11 | 480x360 | 10 | 100 | - | - | - |
- Clone the repository
git clone https://github.com/ntkhoa95/ENet_PyTorch.git
cd ENet_PyTorch
- Download the CamVid datasets
Download the CamVid dataset and unzip to ./content/camvid/
wget https://www.dropbox.com/s/pxcz2wdz04zxocq/CamVid.zip?dl=1 -O CamVid.zip
unzip CamVid.zip- Use command to train the model
python init.py --mode train -iptr ./content/camvid/train/ -lptr ./content/camvid/trainannot/
- Use command to test model
python init.py --mode test -m ./content/checkpoint/camvid/best_model.pth -i ./content/camvid/test/0001TP_008550.png
- Use
--helpto get more commands
python init.py --help
- A. Paszke, A. Chaurasia, S. Kim, and E. Culurciello. Enet: A deep neural network architecture for real-time semantic segmentation. arXiv preprint arXiv:1606.02147, 2016.