Alejandro Newell

The change in the resnet module is based on this paper: https://arxiv.org/pdf/1603.05027.pdf which discusses the effects of switching between pre- and post-activation. Can't remember now if there was much difference...

We mask out the crowd areas in COCO as seen here: https://github.com/umich-vl/pose-ae-train/blob/master/task/loss.py#L16

All detected regions for all images are concatenated into one big list. So: - `im_to_roi_idx` specifies where in the list to find the proposals for a particular image - `num_rois`...

All of the functions you need are in util/eval.lua. Specifically, if you have a set of predictions and their ground truth labels you can first call calcDists() to get normalized...

My recommendation is to set "center" to the center of the bounding box (straightforward enough), and then for "scale" you need to play around a bit. Use the max dimension...

The five values include: the x, y coordinates transformed to the original image space, the x, y values in the heatmap coordinate frame, and then the score of the detection....

The refinement step is useful for COCO but not MPII where the evaluation for COCO does not penalize false positives for individual joints so it helps final performance on the...