Semantic Segmentation Network Based On ResNet38

Introduction

Strictly implement the semantic segmentation network based on ResNet38 of 2018 CVPR PSA(Learning Pixel-level Semantic Affinity with Image-level Supervision for Weakly Supervised Semantic Segmentation). We hope that this code will be of some help to those studying weakly supervised semantic segmentation.

Thanks to the work of jiwoon-ahn, the code of this repository borrow heavly from his AffinityNet repository.

Requirements

• Python 3.6
• pytorch 0.4.1, torchvision 0.2.1
• CUDA 9.0
• 1 x GPUs (11GB)

Usage

Installation

• Download the repository.

git clone https://github.com/peihan-miao/ResNet38-Semantic-Segmentation.git

• Install python dependencies.

pip install -r requirements.txt

• Download model weights from google drive, including ImageNet pretrained models and our training results.

• Download PASCAL VOC 2012 devkit (follow instructions in http://host.robots.ox.ac.uk/pascal/VOC/voc2012/#devkit). It is suggested to make a soft link toward downloaded dataset.

ln -s $your_dataset_path/VOCdevkit/VOC2012 VOC2012

Training and Testing

1. Train SegNet.

python train_seg.py --weights $pretrained_model --image_dir $train_image --mask_dir $train_mask

2. Test SegNet.

python infer_seg.py --weights $pretrained_model --image_dir $test_image_dir --output_crf $output_crf_dir

3. Evaluation.

python evaluation.py --list $VOC2012/ImageSets/Segmentation/[val.txt | train.txt] --predict_dir $your_predict_dir --gt_dir VOC2012/SegmentationClass --comment $your_comments --type npy --curve True

Results and Trained Models

We use the pseudo mask generated by AffinityNet to train the semantic segmentation network and compare it with 2018 CVPR PSA.

Model	Val (mIoU)
ResNet-38	61.7	PSA
ResNet-38	62.4	[Weights]

Reference

[1] J. Ahn and S. Kwak. Learning pixel-level semantic affinity with image-level supervision for weakly supervised semantic segmentation. In Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018.