did anyone test it in realtime?

mks0601 / I2L-MeshNet_RELEASE

Official PyTorch implementation of "I2L-MeshNet: Image-to-Lixel Prediction Network for Accurate 3D Human Pose and Mesh Estimation from a Single RGB Image", ECCV 2020

MIT License

720 stars 128 forks source link

did anyone test it in realtime? #36

Closed Hasankanso closed 3 years ago

Hasankanso commented 3 years ago

I wonder if someone has tested I2L-MeshNet on real time scenario... I did that and I got poor fps and wrong mesh positioning. I tried it actually in a docker container, I've encountered many docker-related problems.

do you have any insight regarding I2L-MeshNet in real time scenario?

mks0601 commented 3 years ago

Did you detect human first using an object detector? The input of I2L-MeshNet is a cropped human image.

ferozkhanabbasi commented 3 years ago

which file to run for the realtime?

Hasankanso commented 3 years ago

which file to run for the realtime?

There's no such file in the repository. You have to integrate an object/person detection model like "yolo" and use it to extract persons, then crop them and feed them to the I2L-MeshNet model. you can put all these in openCV cam loop to get a real-time taste. but I believe it's too slow for real-time. Use demo/demo.py as a start point if you would.

Good luck