yuanyuanli85 / fast_human_pose_estimation_pytorch Goto Github PK

I try to run "python example/mpii.py -a hg --stacks 8 --blocks 1 --checkpoint checkpoint/hg_s8_b1/ " in your code , compare to your log.txt,loss can descend as same with you,but acc samll too mush,just "python example/mpii.py -a hg --stacks 8 --blocks 1 --checkpoint checkpoint/hg_s8_b1/ " is whole?

How to test single image.

hi, does it support multi-person pose estimation?

where is the distillation code? i dont know

when I try to set --mobile=false, it cou;dn't work.

When I try to export like described on the README it gives an error on the model from Google Drive:

python tools/mpii_export_to_onxx.py -a hg -s 2 -b 1 --num-classes 16 --mobile True --in_res 256  --checkpoint checkpoint/mpii_hg_s2_b1_mobile_fpd/model_best.pth.tar --out_onnx checkpoint/mpii_hg_s2_b1_mobile_fpd/model_best.onnx
==> creating model 'hg', stacks=2, blocks=1
=> loading checkpoint 'checkpoint/mpii_hg_s2_b1_mobile_fpd/model_best.pth.tar'
=> loaded checkpoint 'checkpoint/mpii_hg_s2_b1_mobile_fpd/model_best.pth.tar' (epoch 90)
Traceback (most recent call last):
  File "tools/mpii_export_to_onxx.py", line 72, in <module>
    main(parser.parse_args())
  File "tools/mpii_export_to_onxx.py", line 48, in main
    torch.onnx.export(model, dummy_input, args.out_onnx)
  File "/home/ubuntu/anaconda2/envs/fast-human/lib/python2.7/site-packages/torch/onnx/__init__.py", line 25, in export
    return utils.export(*args, **kwargs)
  File "/home/ubuntu/anaconda2/envs/fast-human/lib/python2.7/site-packages/torch/onnx/utils.py", line 131, in export
    strip_doc_string=strip_doc_string)
  File "/home/ubuntu/anaconda2/envs/fast-human/lib/python2.7/site-packages/torch/onnx/utils.py", line 363, in _export
    _retain_param_name, do_constant_folding)
  File "/home/ubuntu/anaconda2/envs/fast-human/lib/python2.7/site-packages/torch/onnx/utils.py", line 278, in _model_to_graph
    _disable_torch_constant_prop=_disable_torch_constant_prop)
  File "/home/ubuntu/anaconda2/envs/fast-human/lib/python2.7/site-packages/torch/onnx/utils.py", line 188, in _optimize_graph
    graph = torch._C._jit_pass_onnx(graph, operator_export_type)
  File "/home/ubuntu/anaconda2/envs/fast-human/lib/python2.7/site-packages/torch/onnx/__init__.py", line 50, in _run_symbolic_function
    return utils._run_symbolic_function(*args, **kwargs)
  File "/home/ubuntu/anaconda2/envs/fast-human/lib/python2.7/site-packages/torch/onnx/utils.py", line 589, in _run_symbolic_function
    return fn(g, *inputs, **attrs)
  File "/home/ubuntu/anaconda2/envs/fast-human/lib/python2.7/site-packages/torch/onnx/symbolic.py", line 130, in wrapper
    args = [_parse_arg(arg, arg_desc) for arg, arg_desc in zip(args, arg_descriptors)]
  File "/home/ubuntu/anaconda2/envs/fast-human/lib/python2.7/site-packages/torch/onnx/symbolic.py", line 90, in _parse_arg
    raise RuntimeError("Failed to export an ONNX attribute, "
RuntimeError: Failed to export an ONNX attribute, since it's not constant, please try to make things (e.g., kernel size) static if possible

sorry to interrupt u. I ran your demo and got an awful result which makes me confused, i'm here to check were you get the same ans pic as i got below or did i do sth wrong？Looking forward to your reply

I am getting the above error (mentioned in title) when i run the code:python tools/mpii_demo.py -a hg -s 2 -b 1 --mobile True --checkpoint checkpoint/mpii_hg_s2_b1_mobile_fpd/model_best.pth.tar --in_res 256 --device cuda .what should I do?

Hi I was hoping if you could shed some light on where I can find the json file for the leeds dataset, I can only find the mpii one?

Thanks,

Hi, I am reproducing your work and think about changing the target student model, such as stack=4, block=2 or stack=4, block=1, and I manually set the num_features and inplanes:

python example/mpii_kd.py -a hg --stacks 4 --blocks 2 --features 32 --inplanes 8 --checkpoint checkpoint/hg_s2b1_f64in8_diqizeng_mobile_fpd --mobile True --teacher_stack 8 --teacher_checkpoint checkpoint/mpii_hg_s8_b1/model_best.pth.tar

this s4b2 model KD training will drop val acc just on 8th epoch from 56% to %17%,

and used:
python example/mpii_kd.py -a hg --stacks 4 --blocks 1 --features 128 --inplanes 32 --checkpoint checkpoint/hg_s2b1_f64in8_diqizeng_mobile_fpd --mobile True --teacher_stack 8 --teacher_checkpoint checkpoint/mpii_hg_s8_b1/model_best.pth.tar

this command will cause val acc drop at 4th epoch from 42% to 1.6%.

But use :
python example/mpii_kd.py -a hg --stacks 2 --blocks 1 --features 64 --inplanes 8 --checkpoint checkpoint/hg_s2b1_f64in8_diqizeng_mobile_fpd --mobile True --teacher_stack 8 --teacher_checkpoint checkpoint/mpii_hg_s8_b1/model_best.pth.tar
It looks normal.

Does the model structure input effect a lot? Do the num_features and inplane have the suitable stack and block structure?

I experiment a lot but still can't achieve the result in the paper, is there any tricks I didn't realize?

If I collect my dataset, I can train it with the code.

Question

Can you provide info about license?
Is it free for commercial purpose? and redis under https://github.com/ilovepose/fast-human-pose-estimation.pytorch/blob/master/LICENSE this license?

we can't visualize validation .

Why do you define two loss function?
total_loss = loss_labeled + unkdloss_alpha * kdloss_unlabeled
total_loss = kdloss_alpha * tsloss + (1 - kdloss_alpha)*gtloss

When I trained the student model under supervision of the teacher model downloaded from the link in readme as well as the labelled data, the validation accuracy drops quite quickly on the 8th epoch, and the best validation accuracy was only about 60%, way low from the paper result. Why does this happen and how to solve this problem? If I train to more epochs will the result be better?

As is mentioned in the paper, the teacher network is choosen as the original Hourglass Network (ECCV 2016), then a student network is trained on a customized network, but the result shows a better perfomance than the result of Newell et al., ECCV’16(the teacher network). I can't understand, could anyone explain?

python example/mpii_kd.py -a hg --stacks 2 --blocks 1 --checkpoint checkpoint/hg_s2_b1_mobile/ mobile=True --teacher_stack 8 --teacher_checkpoint
Hello, does stack 8 mean that the teacher network is 8 hourglasses? Does stacks 2 mean that the student network consists of two hourglass networks?
checkpoint/hg_s8_b1/model_best.pth.tar

Did unlabel dataset need annotation file?
How arrange the unlabel dataset?