HRNet combine CBAM for Cephalometric Landmark Detection

Topdown Heatmap regression based HRNet combine CBAM for Cephalometric Landmark Detection in Lateral X-ray Images

Please focus on the hrnet.py file to see how HRnet+CBAM is implemented, and the configuration file in the configuration file home to set up the tuning parameters for the training loss. After you have downloaded MMPose you can replace the old hrnet.py backbone file in MMPose\mmpose_package\mmpose\mmpose\models\backbones with hrnet.py.

Since this work is based on MMPose and for the CL-detection challenge. Therefore the dataset and some accompanying files need to be downloaded separately.

First install MMPose https://mmpose.readthedocs.io/en/latest/installation.html#installation Get the dataset on the Challenge website https://cl-detection2023.grand-challenge.org/

Data preparation

In this step, you can execute the script step2_prepare_coco_dataset.py . Dont forget change file path.

Model training

Change the path in the configuration file

CUDA_VISIBLE_DEVICES=0 python step3_train_and_evaluation.py
cldetection_configs/td-hm_hrnet-w32_udp-8xb64-250e-512x512_KeypointMSELoss.py
--work-dir='/data/xushuolin/CL-Detection2023/MMPose-checkpoints'

Base on your path change the commend. And run the demo to train.

Test and visualiation

CUDA_VISIBLE_DEVICES=0 python step4_test_and_visualize.py
cldetection_configs/td-hm_hrnet-w32_udp-8xb64-250e-512x512_KeypointMSELoss.py
'/data/zhangHY/CL-Detection2023/MMPose-checkpoints/best_SDR 2.0mm_epoch_40.pth'
--show-dir='/data/xushuolin/CL-Detection2023/MMPose-visualize'

Same step, Base on your path change the commend. And run the demo to test save results about visualiation landmark.