Eldar Insafutdinov

Are you using Tensorflow 1.4? If not, please upgrade.

Sorry for overlooking this, but are you still running into this problem or should I close?

Hi Chris! Thanks for your interest, I'm glad it served you well! It seems like a pretty big change and I would like to look closer at it. I don't...

For testing you need to use different pose_cfg.yaml file, which should look like this one: https://github.com/eldar/pose-tensorflow/blob/master/models/mpii/test/pose_cfg.yaml The test code is not really supported, what does work here is the demo...

Hey, Sure, you can fine-tune from pre-trained model. I just modified pose_cfg.yaml with this line: ``` init_weights: path/to/snapshot-1000000 ``` and started training - it worked. What exactly is wrong for...

I understand now. I think you shouldn't be using the prediction layers of this model, but replace them with your own ones. They should feed from the last layer of...

Hi, this project is not optimized for real time use, though one could get a reasonable speed. First step would be to use ResNet-50 instead of 100-layer version. Part of...

Hi @YYfangzi , Thanks for your interest. I uploaded the training instructions on the MS COCO dataset. Training takes around 7 days on a single GPU. However it could be...

Training on MPII should take less than 3 days, dataset is smaller and also it's preprocessed with tighter crops. On MPII it's 1M iterations.

Hey, the MSCOCO pre-trained model is already available, check out multi-person demo.