Ethan Yanjia Li

but if it's just one or two pixels off the border, we will still have a little bit gaussian remaining for that joint right? because the gaussian patch itself is...

you need to roughly estimate the person height for each image in your custom dataset, and then divide them by 200px.

Thanks for the PR. I didn't have time to fully test this part, and i do remember I added these extra values like `scale` in my other MPII implementation so...

how did you create your TFRecords? specifically, is `image/object/parts/y` field in your TFExample a float number of a integer number?

Do you mean YOLOv3? yes, for the training script itself, you just need to change to `TOTAL_CLASSES=4`. However my implementation uses TF Records so you'll also need to make sure...

seems like the input shape is wrong during the concatenation in the detection head. the part i don't quite understand is that why x and x_medium could have shape of...

hmm there might be some issue with the demo notebook as i didn't make it that generic. what's the raw output of the model?

You should be able to use tools like LabelBox, Scale.a, Playment or FigureEight to label key-points. The MPII dataset I use for Hourglass is a public dataset annotated by some...

Hey John, `tfrecords_mpii.py` is here: https://github.com/ethanyanjiali/deep-vision/tree/master/Datasets/MPII Unlike COCO or ImageNet, MPII is a relatively small dataset, so you should be able to train it quickly. My implementation in this repo...

@DhruvMakwana it looks like the original dataset annotation is off could you check if you have `008664` when you read the original annotation from this part of the code? I...