densebody_pytorch

PyTorch implementation of CloudWalk's recent paper DenseBody.

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

Update on 20190613 A toy dataset has been released to facilitate the reproduction of this project. checkout PREPS.md for details.

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 PREPS.md for details.

• Code reformating complete! Please refer to data_utils/UV_map_generator.py for more details.

• Thanks Raj Advani for providing new hand crafted UV maps!

Training Guidelines

Please follow the instructions PREPS.md to prepare your training dataset and UV maps. Then run train.sh or nohup_train.sh to begin training.

Customizations

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

To explore different network architectures, checkout NETWORKS.md for detailed instructions.

TODO List

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

• [x] [radvani](https://github.com/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] [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] [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.

Authors

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

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

Citation

Please consider citing the following paper if you find this project useful.

DenseBody: Directly Regressing Dense 3D Human Pose and Shape From a Single Color Image

Acknowledgements

The network training part is inspired by BicycleGAN