PyTorch implementation of CloudWalk's recent paper DenseBody.

Note: For most recent updates, please check out the dev branch.

Reproduction results

Here is the reproduction result (left: input image; middle: ground truth UV position map; right: estimated UV position map)


Update Notes

  • SMPL official UV map is now supported! Please checkout for details.
  • Code reformating complete! Please refer to data_utils/ for more details.
  • Thanks Raj Advani for providing new hand crafted UV maps!

Training Guidelines

Please follow the instructions to prepare your training dataset and UV maps. Then run or to begin training.


To train with your own UV map, checkout for detailed instructions.

To explore different network architectures, checkout for detailed instructions.


  • [x] Creating ground truth UV position maps for Human36m dataset.

    • [x] 20190329 Finish UV data processing.
    • [x] 20190331 Align SMPL mesh with input image.
    • [x] 20190404 Data washing: Image resize to 256*256 and 2D annotation compensation.
    • [x] 20190411 Generate and save UV position map.
      • [x] radvani Hand parsed new 3D UV data
      • [x] Validity checked with minor artifacts (see results below)
      • [x] Making UV_map generation module a separate class.
    • [x] 20190413 Prepare ground truth UV maps for washed dataset.
    • [x] 20190417 SMPL official UV map supported!
    • [x] 20190613 A testing toy dataset has been released!
  • [x] Prepare baseline model training

    • [x] 20190414 Network design, configs, trainer and dataloader
    • [x] 20190414 Baseline complete with first-hand results. Something issue still needs to be addressed.
    • [x] 20190420 Testing with different UV maps.


Lingbo Yang(Lotayou): The owner and maintainer of this repo.

Raj Advani(radvani): Provide several hand-crafted UV maps and many constructive feedbacks.