ziwei-zh / corrpm Goto Github PK

Hi you have done a great job
Could you tell me which code you used to generate 16 keypoints like those in Lip.

1. First you may have to do NMS or other post-processing. You may refer to simple baseline repo: https://github.com/microsoft/human-pose-estimation.pytorch/blob/18f1d0fa5b5db7fe08de640610f3fdbdbed8fb2f/lib/core/inference.py#L49
2. I think the upsampling step is not needed since it may cause misalignment issues.

Thank you very much! Finally able to get it work;)

Originally posted by @Sonseca97 in #8 (comment)

dear author ,the dataset link doesnot work ,can you give a new link ? thank you ,my email :[email protected]

Hi, thanks for your code, here are some problem. I can't find LIP_SP_VAL_annotations.json in your project, but run_eval.sh have the code like:
CS_PATH='/data/zzw/segment/data/lip/images_labels' POSE_ANNO='/data/zzw/segment/data/lip/TrainVal_pose_annotations/LIP_SP_VAL_annotations.json'

ModuleNotFoundError: No module named 'libs._ext.__ext'

Interested in trying out.

Dear author:
Thanks for shareing the implementation!
I doubt that How could i test on my own images, Since I dont have the pose and edge info in terms of json file? Thank you.

Hi,
could you suggest a solution to run the evaluation with pytorch 1? I would like to try it on Colab but installing torch 0.4.1 generates problem with Cuda so it is a bit problematic.

The main problem is related to the libs modules but I can't find a complete solution to it :(

It's weird for my PC to train the code, it's very slow, can you figure it out.
In order to use PyTorch:1.4.0, I replace the "InPlaceABNSync" with normal "nn.BatchNorm2d".

CUDA:10.0
PyTorch:1.4.0
RTX 2080ti 11G
Ubuntu 16.04
i7-9700k
RAM：32 GB
iter:0/7616,loss:8.788,lr:1.000e-03,time:8.2 iter:500/7616,loss:3.455,lr:9.997e-04,time:165.6 iter:1000/7616,loss:2.483,lr:9.993e-04,time:166.3 iter:1500/7616,loss:3.580,lr:9.990e-04,time:167.0 iter:2000/7616,loss:2.938,lr:9.987e-04,time:167.5 iter:2500/7616,loss:2.709,lr:9.984e-04,time:166.9 iter:3000/7616,loss:3.379,lr:9.980e-04,time:167.1 iter:3500/7616,loss:2.261,lr:9.977e-04,time:164.4 iter:4000/7616,loss:2.764,lr:9.974e-04,time:164.3 iter:4500/7616,loss:3.147,lr:9.970e-04,time:164.3 iter:5000/7616,loss:3.124,lr:9.967e-04,time:164.3 iter:5500/7616,loss:2.995,lr:9.964e-04,time:164.3 iter:6000/7616,loss:1.715,lr:9.961e-04,time:164.4 iter:6500/7616,loss:2.305,lr:9.957e-04,time:164.0 iter:7000/7616,loss:1.912,lr:9.954e-04,time:165.5 iter:7500/7616,loss:1.704,lr:9.951e-04,time:165.5
Thank you.

Thank for publish your code,

I try to run ./run_train.sh, but face this error. I cannot find this files in utils folder. Does your class CriterionPoseEdge different from class CriterionAll in CE2P repo?

Traceback (most recent call last):
File "train.py", line 24, in
from utils.criterion import CriterionPoseEdge
ImportError: cannot import name 'CriterionPoseEdge'

Dear author ,thanks your code.About 16 pose point ,I have a question,how to get this result.I get this parameter use blew code：
points=outputs[2][0]
points = interp(points).data.cpu().numpy()
points = points.transpose(0, 2, 3, 1)
points_preds[idx:idx + num_images, :, :] = np.asarray(np.argmax(points, axis=3), dtype=np.uint8)
point_pred=np.asarray(np.argmax(points, axis=3), dtype=np.uint8)
but how to map this shape (116 96*96) map orign image,hope yor answer.thank you @ziwei-zh

I think you have done a great work. Could you tell me, where could download Mask R-CNN that you used in your paper?

Hi, can you tell me how to save pose and edge predictions as images? Thank you very much.

I also encountered the issue as well as liutinglt/CE2P#34.
However, this issue does not provide a solution.
I do not change the part of weight loading and it does not output any error messages.
How do I fix it?
Note that I referred to the CE2P repository (https://github.com/liutinglt/CE2P) so that exchange ABN for PyTorch ver 0.x to ABN for PyTorch ver 1.x.

The output of the network goes from about -0.02 to 0.03. Is it collect?
The network generates a poor result though I exchanged ABN to BN.
Thanks.

--My environment--
GPU: Tesla V100
PyTorch: 1.6
CUDA: 10.2.89
cuDNN: 8.0.2
ABN: https://github.com/mapillary/inplace_abn
Pre-trained model: LIP_best.pth

Hi, may I ask if the model is able to generate pose coordinates as output on a test image? I found the model output is a list where you only used seg2. However, when I use the output of pose1, I get the shape of 12x16x96x96 where 12 is the batch size and 16 is the number of keypoints labels. I am not sure how to proceed from this stage. Hope you can help me clarify! Thank you!
return [[seg1, seg2], [edge1], [pose1]]

ModuleNotFoundError: No module named 'libs._ext.__ext'