Skip to content

Code repo for the paper "Semantic Correspondence via 2D-3D-2D Cycle"

Notifications You must be signed in to change notification settings

qq456cvb/SemanticTransfer

Repository files navigation

SemanticTransfer

Code repo for the paper Semantic Correspondence via 2D-3D-2D Cycle.

Demo

Please run demo.py.

Pretrained Weights

You can download them from Google Drive.

Training

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 mitsuba and generate their corresponding viewpoints through preprocess.py.
  • Train viewpoint estimation network by running scripts/train_vp.sh.
  • Train 3D embedding prediction network by running train_embs.py and then generate keypoints' average embeddings for retrieval. This step requires KeypointNet dataset.

About

Code repo for the paper "Semantic Correspondence via 2D-3D-2D Cycle"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published