🥧 PI3DETR: Detection of Sharp 3D CAD Edges [CPU-PREVIEW]

A novel end-to-end deep learning model for parametric curve inference in 3D point clouds and meshes.

🧩 Supported Inputs

  • Point Clouds: .xyz, .ply; Meshes: .obj
  • Mesh is surface-sampled using Max Points (display) slider.

⚙️ Point Cloud Settings

  • Adjust Max Points, point size, and axes visibility.
  • Controls visualization of point cloud.

🎯 Confidence Thresholds

  • Hover to inspect scores.
  • Filter curves by class confidence interactively

🧠 Model Settings

  • Sampling Mode: Choose downsampling strategy.
  • Model Input Size: Number of model input points.
  • Queries: Transformer decoder queries (max. output curves).

⚡ Performance Notes

  • Trained on human-made objects.
  • Optimized for GPU; this demo runs on CPU.
  • For full qualitative performance: GitHub → PI3DETR

▶️ Run Inference

  • Click on demo point clouds (from test set) below.
  • Press Run PI3DETR to execute inference and visualize results.

Point Cloud Settings
0 500000
1 8

Model Settings
Main Sampling Method
1000 100000
32 512

Confidence Thresholds (per class)
0 1
0 1
0 1
0 1

Demo Point Clouds (click an image to load)