deep-high-resolution-net.pytorch
deep-high-resolution-net.pytorch copied to clipboard
leoxiaobin
how to get 'scale' in mpii dataset?
The authors provide json files for MPII dataset (train.json, trainval.json..).
I want to create json files for my custom dataset to use the code for MPII dataset.
but I do not find the specific description about that, especially 'scale'.
how to get 'scale' value?
following is an example in valid.json that they provided.
{
"joints_vis": [
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1
],
"joints": [
[
804.0,
711.0
],
[
816.0,
510.0
],
[
908.0,
438.0
],
[
1040.0,
454.0
],
[
906.0,
528.0
],
[
883.0,
707.0
],
[
974.0,
446.0
],
[
985.0,
253.0
],
[
982.7591,
235.9694
],
[
962.2409,
80.0306
],
[
869.0,
214.0
],
[
798.0,
340.0
],
[
902.0,
253.0
],
[
1067.0,
253.0
],
[
1167.0,
353.0
],
[
1142.0,
478.0
]
],
"image": "005808361.jpg",
"scale": 4.718488,
"center": [
966.0,
340.0
]
}
The authors provide json files for MPII dataset (train.json, trainval.json..).
I want to create json files for my custom dataset to use the code for MPII dataset.
but I do not find the specific description about that, especially 'scale'.
how to get 'scale' value?
following is an example in valid.json that they provided.
{ "joints_vis": [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ], "joints": [ [ 804.0, 711.0 ], [ 816.0, 510.0 ], [ 908.0, 438.0 ], [ 1040.0, 454.0 ], [ 906.0, 528.0 ], [ 883.0, 707.0 ], [ 974.0, 446.0 ], [ 985.0, 253.0 ], [ 982.7591, 235.9694 ], [ 962.2409, 80.0306 ], [ 869.0, 214.0 ], [ 798.0, 340.0 ], [ 902.0, 253.0 ], [ 1067.0, 253.0 ], [ 1167.0, 353.0 ], [ 1142.0, 478.0 ] ], "image": "005808361.jpg", "scale": 4.718488, "center": [ 966.0, 340.0 ] }
Maybe you can get the scale array from the function named "_xywh2cs" in Simple-HRNet/datasets/COCO.py. You need obtain the bbox of your dataset first.
