search

isarandi / metrabs

Estimate absolute 3D human poses from RGB images.
https://arxiv.org/abs/2007.07227
MIT License
441 stars 65 forks source link

GradCam #43

Open chrisdottel opened 1 year ago

chrisdottel commented 1 year ago

I'd love to try and implement GradCam on this project, however the pretrained models seem to be packed with the Tensorflow save instead of Keras. Do you have the checkpoints available for the evaluation or training code for finetuning/ custom implementations?

isarandi commented 1 year ago

Do you necessarily need a Keras-style checkpoint for this? It should be possible to backprop through the TF SavedModel as well and get gradients wrt the input image. You may want to use the model.crop_model for this analysis, which is the single-person pose estimator at the core of the method, which takes a 256x256 cropped image. (See bottom of https://github.com/isarandi/metrabs/blob/master/docs/INFERENCE_GUIDE.md)