A Keras implementation of Pyramid Scene Parsing Networks PSPNet

Original Paper and implementation:

@inproceedings{zhao2017pspnet,
  author = {Hengshuang Zhao and
            Jianping Shi and
            Xiaojuan Qi and
            Xiaogang Wang and
            Jiaya Jia},
  title = {Pyramid Scene Parsing Network},
  booktitle = {Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2017}
}

Installation:

python3 setup.py install

Usage:

python3 pspnet.py
python3 pspnet.py -m pspnet101_cityscapes -i ../example_images/munster_000013_000019_leftImg8bit.png -o ../example_results/munster_000013_000019_leftImg8bit.png -s -f
python3 pspnet.py -m pspnet101_cityscapes -i $CITYSCAPES_DATASET/leftImg8bit/val -o $CITYSCAPES_RESULTS -s -ms -f
python3 pspnet.py -m pspnet101_voc2012 -i ../example_images/000129.jpg -o ../example_results/000129.png -s -f
python3 pspnet.py -m pspnet50_ade20k -i ../example_images/ADE_val_00000435.jpg -o ../example_results/ADE_val_00000435.png -s -f
python3 pspnet.py -m pspnet50_ade20k -i ../example_images/ade20k.jpg -o ../example_results/ade20k.png -s -f

Results:

ADE20K:

CityScapes

Pascal Voc 2012

Converting caffe weights

Converted trained weights are needed to run the network and will be downloaded from dropbox the first time you use a model. The weights of the original caffemodel were converted with weight_converter.py as follows:

python3 weight_converter.py <path to .prototxt> <path to .caffemodel>

Running the converter needs the compiled original PSPNet caffe code and pycaffe.