Densepose on video

Put the .py files with your detectron installation.

Use command

python2 tools/infer_vid.py     
--cfg configs/DensePose_ResNet101_FPN_s1x-e2e.yaml     
--output-dir DensePoseData/infer_out/    
--wts https://s3.amazonaws.com/densepose/DensePose_ResNet101_FPN_s1x-e2e.pkl 
--input-file filename

Note: Requires about 4GB VRAM.

If you use ResNet101_FPN_s1x model, you need more than 4GB of memory. If you have only 4GB use ResNet50_FPN_s1x.

You can get configs and models from here.

• https://github.com/facebookresearch/DensePose/tree/master/configs

• https://github.com/facebookresearch/DensePose/blob/master/MODEL_ZOO.md

Additional requirement - ffmpeg

Tested with

• Ubuntu 16.04

• Cuda 9.0

• CUDNN 6

• opencv 3.4

• GTX 1050 Ti

• GTX 1080 Ti

Example, ran on Geralt doing some witcherin'

And a bit of Night city.