Skip to content

This is an ongoing re-implementation of DeepLab_v3_plus on pytorch which is trained on VOC2012 and use ResNet101 for backbone.

License

Notifications You must be signed in to change notification settings

songdejia/DeepLab_v3_plus

Folders and files

NameName
Last commit message
Last commit date

Latest commit

7822d94 · Nov 23, 2018

History

37 Commits
Oct 23, 2018
Oct 23, 2018
Oct 23, 2018
Nov 23, 2018
Oct 23, 2018
Nov 5, 2018
Oct 23, 2018
Oct 23, 2018
Oct 23, 2018
Oct 23, 2018

Repository files navigation

DeepLab_V3_plus : a model about semantic segmentation

This is a simple pytorch re-implementation of Google Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation.

Introduction:

This work still need to be updated. The features are summarized blow:

  • Use ResNet101 as base Network. Xception will be updated soon.
  • Use only VOC2012 for base dataset. Other dataset will be updated soon.

We have finished:

  • Version avaliable for VOC2012.

  • You can check your dataloader error in 'path/to/workspace/check/check_dataloader/img'. You will see three part:

    • 1.original image which we load directly from image path.
    • 2.restore image from torch-tensor(transformed from numpy) back to numpy ndarray.
    • 3.mask loaded by dataloader. Following images show original image, augumentation image and mask target from left to right
  • 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.

Usage:

  • Download dataset and unzip
ln -s VOCdevkit path/to/deeplab_v3_plus/dataset
  • Pretrained model is avaliable BaiduYun Link
  • Finally, run the model.
  • Check you GPU resources and modify your run.sh.
sh run.sh

Future:

  • Xception as Network Baseline.
  • Pretrained model on COCO, JFT.
  • Depthwise separable convolution.
  • Support dataset for Cityscapes.
  • Visualization for test result and gt.

About

This is an ongoing re-implementation of DeepLab_v3_plus on pytorch which is trained on VOC2012 and use ResNet101 for backbone.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published