Pytorch implementation of FCN8 and PSPNet

Download trained Deeplab Resnet model from: https://drive.google.com/open?id=0BxhUwxvLPO7TeXFNQ3YzcGI4Rjg

FCN8 and PSPNet were trained on the PASCAL VOC 2012 Dataset and the BRATS 2017 Dataset.

Results:

FCN8:

Validation: Pixel accuracy: 83%, Mean IU: 43%

PSPNet:

Validation: Pixel accuracy: 82%, Mean IU: 62%

Setup: Install the environment using:

conda env create -f environment.yml

Activate the environment:

source activate psp_env

Start visdom on port 8097:

visdom

Training:

python main.py [-h] --dataset DATASET --model MODEL [--train] [--test]
                    [--checkpoint CHECKPOINT] --config CONFIG
                    [--exp_suffix EXP_SUFFIX] [--device {cpu,cuda:0,cuda:1}]
                    [--image IMAGE] [--label LABEL]

Example:

python main.py --dataset VOC12 --model PSPNet --train --config ../configs/config.ini --exp_suffix 1 --device cpu

Validation:

python main.py --dataset VOC12 --model PSPNet --test --checkpoint <checkpoint_file>  --config ../configs/config.ini --exp_suffix 1 --image input_image --label output_image

References:

https://github.com/bodokaiser/piwise

https://github.com/isht7/pytorch-deeplab-resnet