anibali / margipose

3D monocular human pose estimation
Apache License 2.0
98 stars 20 forks source link

'margipose.data.mpi_inf_3dhp', 'sequence_info.json' #31

Closed sunmengnan closed 2 years ago

sunmengnan commented 2 years ago

could you upload the files above extracted from "mpii_get_sequence_info.m", i cannot get these files due to matlab limatations, thx.

anibali commented 2 years ago

That file is a part of the repository, you can find it here: https://github.com/anibali/margipose/blob/master/src/margipose/data/mpi_inf_3dhp/sequence_info.json

sunmengnan commented 2 years ago

in Constants = {

Training set sequences: (subject_id, sequence_id)

'train_seqs': [
    (1, 1), (1, 2), (2, 1), (2, 2), (3, 1), (3, 2), (4, 2),
    (5, 1), (5, 2), (6, 1), (6, 2), (7, 1), (7, 2), (8, 1),
],
# Validation set sequences: (subject_id, sequence_id)
'val_seqs': [
    (4, 1), (8, 2)
],
# Camera IDs used for training/validation sets (same as VNect)
'vnect_cameras': [0, 1, 2, 4, 5, 6, 7, 8],
# Total number of cameras
'n_cameras': 14,
# Per-sequence information extracted from "mpii_get_sequence_info.m"
'seq_info': json.loads(read_text('margipose.data.mpi_inf_3dhp', 'sequence_info.json')),
# Root joint index (pelvis) for training/validation data
'root_joint': 4,
# Videos with known problems
'blacklist': {
    'S6/Seq2': [2],  # imageSequence/video_2.avi is too short
},

} only S6/Seq2 in the blacklist, is it only for test?

anibali commented 2 years ago

S6/Seq2 is a training sequence, but there's something wrong with the video_2.avi file so that particular view is skipped for that sequence when preprocessing the dataset.