This is an ongoing re-implementation of DeepLab_v3_plus on pytorch which is trained on VOC2012 and use ResNet101 for backbone.
This is a simple pytorch re-implementation of Google Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation.
This work still need to be updated. The features are summarized blow:
Version avaliable for VOC2012.
You can check your dataloader error in 'path/to/workspace/check/check_dataloader/img'. You will see three part:
Network architecture.
Evaluation mIOU on PASCAL VOC2012 valset every epoches.
Pretrained model on PASCAL VOC2012 BaiduYun Link, which is trained on 2 Tesla P100 for 100 epoches with config shown in code.
ln -s VOCdevkit path/to/deeplab_v3_plus/dataset
sh run.sh