search

Walter0807 / MotionBERT

[ICCV 2023] PyTorch Implementation of "MotionBERT: A Unified Perspective on Learning Human Motion Representations"
Apache License 2.0
1.08k stars 135 forks source link

Json format for use with 3D Pose motion bert #145

Open CrisTTT opened 1 month ago

CrisTTT commented 1 month ago

Hi, I'm trying to inference motion bert in the wild. I am having issues with the JSON format that I should use for this.

Below is the json format i am currently using. I can't find an example of a good JSON used for this in order to reformat it to the correct format.

{ "annotations": [ { "frame_id": 0, "keypoints": [ 256.0015869140625, 455.171630859375, 0.0, 258.3621826171875, 452.07177734375, 0.0, 264.28759765625, 448.86700439453125, 0.0, 269.7962646484375, 457.30731201171875, 0.0, 301.1651611328125, 450.5689697265625, 0.0, 272.2452392578125, 496.71942138671875, 0.0, 332.996337890625, 473.644287109375, 0.0, 259.823974609375, 527.5805053710938, 0.0, 360.90948486328125, 497.08831787109375, 0.0, 254.77301025390625, 496.04486083984375, 0.0, 375.521484375, 638.1166381835938, 0.0, 320.78118896484375, 609.2880249023438, 0.0, 365.06146240234375, 593.7268676757812, 0.0, 323.9300537109375, 698.5770874023438, 0.0, 357.780517578125, 632.9181518554688, 0.0, 329.4959716796875, 797.6443481445312, 0.0, 388.8597412109375, 688.8500366210938, 0.0 ], "bbox": [ 0, 0, 478, 850 ] }, { "frame_id": 1, "keypoints": [ 257.859619140625, 454.60906982421875, 0.0, 259.45166015625, 451.2744140625, 0.0, 266.6239013671875, 447.96405029296875, 0.0, 270.135498046875, 457.2606201171875, 0.0, 302.27679443359375, 450.3916015625, 0.0, 273.5531005859375, 496.06805419921875, 0.0, 333.98193359375, 473.67254638671875, 0.0, 259.32861328125, 526.0870971679688, 0.0, 361.90771484375, 499.54449462890625, 0.0, 254.3492431640625, 494.46270751953125, 0.0, 381.62225341796875, 648.5664672851562, 0.0, 320.60833740234375, 609.2427368164062, 0.0, 364.91314697265625, 594.4885864257812, 0.0, 324.10009765625, 698.9760131835938, 0.0, 356.1573486328125, 631.1774291992188, 0.0, 327.9652099609375, 795.2775268554688, 0.0, 389.0799560546875, 689.3659057617188, 0.0 ], "bbox": [ 0, 0, 478, 850 ] },

CrisTTT commented 1 month ago

vis_climb-trim.json Here is an example of a JSON I want to use. The error I am encountering for: python infer_wild.py --vid_path climb-trim.mp4 --json_path vis_climb-trim.mp4.json --out_path output_dump

is the following: Traceback (most recent call last): File "infer_wild.py", line 61, in wild_dataset = WildDetDataset(opts.json_path, clip_len=opts.clip_len, scale_range=[1,1], focus=opts.focus) File "D:\motion\motionbert\lib\data\dataset_wild.py", line 95, in init self.vid_all = read_input(json_path, vid_size, scale_range, focus) File "D:\motion\motionbert\lib\data\dataset_wild.py", line 80, in read_input kpts_all = halpe2h36m(kpts_all) File "D:\motion\motionbert\lib\data\dataset_wild.py", line 51, in halpe2h36m raise ValueError("Input keypoints do not match the expected format. Expected at least 20 keypoints.") ValueError: Input keypoints do not match the expected format. Expected at least 20 keypoints.