[NeurIPS 2024] Key-Grid: Unsupervised 3D Keypoints Detection using Grid Heatmap Features

This repository provides the code for paper:
Key-Grid: Unsupervised 3D Keypoints Detection using Grid Heatmap Features

💻 Installation

• ClothesNet dataset
• PyTorch ≥ 1.4
• Scikit-learn
• Open3d
• Pyvista
• Pytorch3d

🛠️ Usage

Train the network

Train the pointnet++ on the train dataset. Here, we take the chair category as an example. Note, if you want to train other categories, you should change the dataset root.

python train.py 

Predict the keypoint

Predict the keypoint on the test dataset. We get the keypoint predicted by the pointnet++ on the ClothesNet dataset.

python predict_keypoint.py  

🔍 Visualizer

In this section, we provide code to visualize keypoints and point cloud. 1.Predict the keypoints on the test dataset. You should run predict_keypoint.py; 2. Visualization results of keypoints and point cloud. You should run vision.py.

😺 Acknowledgement

This code inherits some codes from Skeleton Merger, SC3K.