Introduction

Our model comes from the MMSegmentation library. In order to use MMsegmentation, please follow the official tutorial : https://github.com/open-mmlab/mmsegmentation#readme

Download

Dataset

• Stanford Healthcare provides an open source cardiac echocardiography dataset, containing 10,030 apical-4-chamber echocardiography videos. The dataset is publicly available at https://echonet.github.io/dynamic/

• Based on the video dataset, we proposed a dataset of 1047 images, each image was manually annotated left and right ventricles. The dataset is available here

• Download the dataset and extract it into ./data/, without creating subfolders.

Pretrained models

• Pretrained models can be downloaded here. To reproduce the results, the pretrained models(*.pth) should be placed in ./pretrained/, and then test model directly

Run code

Training

• Train the deeplabv3_unet_s5-d16 model:

python tools/train.py configs/unet/deeplabv3_unet_s5-d16_128x128_40k_Car_0505.py --work-dir=pretrained/2class_cardiac

Testing

• For testing, run:

python tools/test.py configs/unet/deeplabv3_unet_s5-d16_128x128_40k_Car_0505.py\
    pretrained/2class_cardiac/latest.pth \
    --show-dir results

• To reproduce the results, move the pretrained models(*.pth) downloaded from here to ./pretrained/, and then test model directly.

Other scripts

Calculate mean & std

• Calculate the mean and std of custom dataset:

python ./get_mean.py ${img_pth} 

Save video

• Save the output image sequence as a video:

python ./save_video.py ${original_pth} ${target_pth}