DeepLab To Segment water in flood images

This is an unofficial PyTorch modification of DeepLab v2 with a ResNet-101 backbone. Most of the code has been taken from this repository: kazuto1011/deeplab-pytorch

We use their model trained on COCO-Stuff 164K dataset and merged some labels to output a binary mask of water.

Merged labels:

water-other
river
sand
sea

For more info on how we selected this model one should look at this thread: cc-ai/kdb/issues/62

Performance

Evaluating on our dataset of manually annotated 210 images of flood, without the CRF post processing, results appear very good, the mean IOU is about 79% with some example that totally fail.

A good way of using it would be to process our entire dataset and remove manually every image with less than 50% of water covered. If we do that post evaluation on the 210 images, we keep 88 % of them and increase the IOU to 86%. Further filtering could be done..

Here is an example of a good segmentation.

CRF should increase the performance but is slower at inference time (9s/it vs several it/s).

Pretrained Model

Pretrained model is provided here: deeplabv2_resnet101_msc-cocostuff164k

Requirements

Python 2.7+/3.6+
Anaconda environement

Then setup from conda_env.yaml. Please modify cuda option as needed (default: cudatoolkit=10.0)

$ conda env create -f configs/conda_env.yaml
$ conda activate deeplab-pytorch

How to use ?

Single image

Usage: demo.py single [OPTIONS]

  Inference from a single image

Options:
  -c, --config-path FILENAME  Dataset configuration file in YAML  [required]
  -m, --model-path PATH       PyTorch model to be loaded  [required]
  -i, --image-path PATH       Image to be processed  [required]
  --cuda / --cpu              Enable CUDA if available [default: --cuda]
  --crf                       CRF post-processing  [default: False]
  --help                      Show this message and exit.

Several images

Usage: demo.py test [OPTIONS]

  Inference from a single folder

Options:
  -c, --config-path FILENAME  Dataset configuration file in YAML  [required]
  -m, --model-path PATH       PyTorch model to be loaded  [required]
  -i, --folder-path PATH      Folder of images to be processed  [required]
  -o, --output-path PATH      Folder to store results [required]
  --cuda / --cpu              Enable CUDA if available [default: --cuda]
  --crf                       CRF post-processing  [default: False]
  --help                      Show this message and exit.

Example:

python demo.py test -c ./configs/cocostuff164k.yaml -m ./deeplabv2_resnet101_msc-cocostuff164k-100000.pth -i ./Dataset/flood_dataset/ -o ./output/predicted_crf/ --crf

References

L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, A. L. Yuille. DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs. IEEE TPAMI, 2018.
Project / Code / arXiv paper
H. Caesar, J. Uijlings, V. Ferrari. COCO-Stuff: Thing and Stuff Classes in Context. In CVPR, 2018.
Project / arXiv paper
M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, A. Zisserman. The PASCAL Visual Object Classes (VOC) Challenge. IJCV, 2010.
Project / Paper

cc-ai/deeplab-pytorch