About BodyPoseEstimator
yangfeixia88 opened this issue · 4 comments
Thanks! You need to install the OpenPose API wrapper for this.
pip install git+https://github.com/prasunroy/openpose-pytorch.git
For more information check https://github.com/prasunroy/openpose-pytorch.
If you are training from scratch then simply run python train.py
. However, if you like to resume training from a previous checkpoint, then you can specify the existing directory containing the checkpoints and the respective checkpoint number.
For example, to resume training from the pretrained checkpoint at iteration 260500 modify lines 27-28
of train.py
as follows:
ckpt_id = 260500
ckpt_dir = f'../output/{dataset_name}/ckpt/pretrained'
Every run of train.py
will create a new directory under f'../output/{dataset_name}/ckpt/
for saving checkpoints for that run. The name of such a directory is the run_id
for that specific run. This run_id
will be used in both test.py
and eval.py
for testing and evaluation respectively.
During testing, you need to specify the run_id
and ckpt_ids
for a specific run.
For example, to test the pretrained checkpoints at iterations 257500 and 260500 modify lines 22-23
of test.py
as follows:
run_id = 'pretrained'
ckpt_ids = [257500, 260500]
Then simply run python test.py
.
Similarly, during evaluation, you need to specify the run_id
and ckpt_ids
for a specific run. However, make sure to run test.py
first to generate the images for the specific checkpoint(s) you wish to evaluate.
For example, to evaluate the pretrained checkpoint at iteration 257500 modify lines 19-20
of eval.py
as follows:
run_id = 'pretrained'
ckpt_ids = [257500]
Then simply run python eval.py
.
Thank you very much for your patient reply and your excellent work.