Now that I know that the pkl file is the source location of the ground truth 2D poses corresponding to the input images,
I would like to ask four additional things:
Does the pkl file contain 2D poses of only subset of Human 3.6M validation images? Or does in contain 2D poses of WHOLE set of Human 3.6M validation set?
If the pkl file contains 2D poses of only subset of Human 3.6M validation images, what is the list of images in the subset?
When I split the huge pkl file using tools/convert_h36m.py, it results in smaller individual pkl files with names 00000000.pkl, 00000001.pkl, 00000002.pkl, and so on. To what Human3.6M dataset image does each of these pkl files correspond to?
Are the 2D pose values in the pkl file ground truths? Or are they just detections from AlphaPose? If they are detections from AlphaPose, did the AlphaPose detections make use of ground truth bounding boxes? Or did they make use of bounding boxes detected by some off-the-shelf detectors like YOLOv3?
Thank you for your answer to https://github.com/Walter0807/MotionBERT/issues/64
Now that I know that the pkl file is the source location of the ground truth 2D poses corresponding to the input images, I would like to ask four additional things:
tools/convert_h36m.py
, it results in smaller individual pkl files with names00000000.pkl
,00000001.pkl
,00000002.pkl
, and so on. To what Human3.6M dataset image does each of these pkl files correspond to?Thank you very much in advance.