Code repo for the paper Semantic Correspondence via 2D-3D-2D Cycle.
You can download them from Google Drive.
Training the full pipeline is somewhat involved and complicated, and our code is heavily based on ShapeHD. In general, there are four steps:
- Train ShapeHD model as outlined here.
- Prepare synthetic ShapeNet model renderings by
mitsubaand generate their corresponding viewpoints through
- Train viewpoint estimation network by running
- Train 3D embedding prediction network by running
train_embs.pyand then generate keypoints' average embeddings for retrieval. This step requires KeypointNet dataset.