Repository files navigation

2021-10-25: This repo is no longer under maintenance. My apologies, should have done this a long time ago.

This is a tensorflow implementation for paperDense Human Body Correspondences Using Convolutional Networks.

ATTENTION: This repo is currently semi-finished, in next few weeks, the newest:

• visualize scripts
• training tutorial

will be updated.

1. Clone this repository to your computer.
2. Modifyproject_dir inconfig.py to the path of this repo.
3. Download 3D human model meshesdata.zip (48M), unzip it to the repo directory. The structure should be likepath/to/repo/data/...
4. Download pretrained modelalex-SM-5 (121M), unzip it to the models directory. The structure should be likepath/to/repo/models/alex-SM-5/..

For input depth image with shape [H, W, 1], outputs [H, W, 16] numpy array. Example:

python predict.py --checkpoint ./models/alex-SM-5/model --output feature.npy --depth ./test.png

For input human mesh model (supportobj andply), outputs [vertex_count, 16] numpy array.The input mesh will be format as 1.8m tall, zero-centerd.

python predict.py --checkpoint ./models/alex-SM-5/model --output feature.npy --mesh ./test.obj --flipyz