In this paper, we present "A Self-supervised Pressure Map Keypoint Detection Approach: Balancing Generalization and Computational Efficiency across Tasks and Datasets." Our proposed method, referred to as Self-supervised Pressure Map Keypoint Detection (SPMKD), is specifically designed to accommodate the unique image characteristics of pressure maps. The SPMKD approach comprises two main components: an Encoder-Fuser-Decoder (EFD) model, and a Classification-to-Regression Weight Transfer (CRWT) technique. When evaluated on the SLP and SMaL datasets, our model demonstrates superior computational efficiency and generalization performance compared to conventional approaches.
- SLP dataset: https://github.com/ostadabbas/SLP-Dataset-and-Code
- SMaL dataset: https://ieee-dataport.org/documents/simultaneously-collected-multimodal-mannequin-lying-pose-smal
As the weights file exceeds the Github limit, we use anonymity to share the weights we have trained. All you need to do is download the weights folder and place it in the root of your project.
- ResNet
python main.py --phase test --detection_head ResNet - DenseNet
python main.py --phase test --detection_head DenseNet - SPMKD+ResNet
python main.py --phase test --detection_head ResNet --use_Encoder - SPMKD+DenseNet
python main.py --phase test --detection_head DenseNet --use_Encoder
- SPMKD+GCN+SLP
python main.py --phase test --use_Keypoint --keypoint_type Ours --GNN_Network GCN - SPMKD+GAT+SLP
python main.py --phase test --use_Keypoint --keypoint_type Ours --GNN_Network GAT - SPMKD+GraphSAGE+SLP
python main.py --phase test --use_Keypoint --keypoint_type Ours --GNN_Network GraphSAGE
- SPMKD+GCN+SMaL
python main.py --phase test --use_Keypoint --keypoint_type Ours --GNN_Network GCN --SMaL - SPMKD+GAT+SMaL
python main.py --phase test --use_Keypoint --keypoint_type Ours --GNN_Network GAT --SMaL - SPMKD+GraphSAGE+SMaL
python main.py --phase test --use_Keypoint --keypoint_type Ours --GNN_Network GraphSAGE --SMaL