menghao666 / minimal-hand-pytorch Goto Github PK

Hey,
I tried installing the environment, unfortunately, it fails either due to manopth or the MANO models.
Can you help me out? Anything I need to specify differently?

(minimal-hand-torch) muellelu@e130-pc25:~/git/minimal-hand-pytorch$ python demo.py
Traceback (most recent call last):
  File "demo.py", line 3, in <module>
    from manopth import manolayer
  File "/home/AD/muellelu/xconda3/envs/minimal-hand-torch/lib/python3.7/site-packages/manopth/manolayer.py", line 7, in <module>
    from mano.webuser.smpl_handpca_wrapper_HAND_only import ready_arguments
  File "/home/AD/muellelu/git/minimal-hand-pytorch/mano/webuser/smpl_handpca_wrapper_HAND_only.py", line 137
    print 'FINITO'
                 ^
SyntaxError: Missing parentheses in call to 'print'. Did you mean print('FINITO')?
(minimal-hand-torch) muellelu@e130-pc25:~/git/minimal-hand-pytorch$ 

hi~have you ever try some other more efficient network which can be run on mobile phone?

An issue from the official repo.

Hi, I am wondering, once the pose and shape parameters are estimated, the absolute scale of the hand model will also be recovered. Does it mean that the network estimates an absolute-scale hand model from a single image?
I am a new hand in this area, is the MANO model scale-independent?

I find all the link assocaited with STB datasets are miss. Could you provide the link could download STB datasets?

Hi, I found that the solver = LM_solver... in demo.py do not be used, and the solver in op_pso.py is not define.

我不能仅仅用“cmu”数据集去train，在hand143_panopticdb.py中，我发现他们不能正确的使用3D数据
raceback (most recent call last):
File "/ssd3/hfang/Mini-hand2/train_detnet.py", line 547, in
main(parser.parse_args())
File "/ssd3/hfang/Mini-hand2/train_detnet.py", line 160, in main
args=args, loss_all=loss_all
File "/ssd3/hfang/Mini-hand2/train_detnet.py", line 348, in train
am_loss_dm.update(losses['det_dm'].item(), targets['batch_3d_size'].item())
File "/ssd3/hfang/Mini-hand2/utils/eval/evalutils.py", line 26, in update
self.avg = self.sum / self.count
ZeroDivisionError: float division by zero

the file (rhd_pso.npy) is not find.I didn't find any pattern files,for example do_pso.npy,eo_pso.npy.stb_pso.npy 。

Hi!
When I tried using the command 'conda env create --file=environment.yml', it threw me the following error:

Collecting package metadata (repodata.json): done
Solving environment: failed

ResolvePackageNotFound:

• pcre==8.44=he6710b0_0
• pixman==0.40.0=h7b6447c_0
• tornado==6.1=py37h27cfd23_0
• libopencv==3.4.2=hb342d67_1
• ld_impl_linux-64==2.33.1=h53a641e_7
• qt==5.9.7=h5867ecd_1
• libgfortran-ng==7.3.0=hdf63c60_0
• freetype==2.10.4=h5ab3b9f_0
• pandas==1.2.2=py37ha9443f7_0
• sip==4.19.8=py37hf484d3e_0
• libopus==1.3.1=h7b6447c_0
• numpy-base==1.19.2=py37hfa32c7d_0
• cffi==1.14.5=py37h261ae71_0
• libxcb==1.14=h7b6447c_0
• zstd==1.4.5=h9ceee32_0
• readline==8.1=h27cfd23_0
• libuv==1.40.0=h7b6447c_0
• pyqt==5.9.2=py37h05f1152_2
• harfbuzz==1.8.8=hffaf4a1_0
• libuuid==1.0.3=h1bed415_2
• gstreamer==1.14.0=h28cd5cc_2
• torchvision==0.4.0=py37_cu100
• libffi==3.3=he6710b0_2
• ca-certificates==2021.1.19=h06a4308_0
• lz4-c==1.9.3=h2531618_0
• mkl_fft==1.3.0=py37h54f3939_0
• expat==2.2.10=he6710b0_2
• libtiff==4.1.0=h2733197_1
• libgcc-ng==9.1.0=hdf63c60_0
• libstdcxx-ng==9.1.0=hdf63c60_0
• xz==5.2.5=h7b6447c_0
• icu==58.2=he6710b0_3
• glib==2.67.4=h36276a3_1
• setuptools==52.0.0=py37h06a4308_0
• pillow==8.1.0=py37he98fc37_0
• termcolor==1.1.0=py37h06a4308_1
• zlib==1.2.11=h7b6447c_3
• libpng==1.6.37=hbc83047_0
• python==3.7.10=hdb3f193_0
• mkl==2020.2=256
• scipy==1.6.1=py37h91f5cce_0
• kiwisolver==1.3.1=py37h2531618_0
• jasper==2.0.14=h07fcdf6_1
• ncurses==6.2=he6710b0_1
• mkl-service==2.3.0=py37he8ac12f_0
• numpy==1.19.2=py37h54aff64_0
• libxml2==2.9.10=hb55368b_3
• pytorch==1.2.0=py3.7_cuda10.0.130_cudnn7.6.2_0
• hdf5==1.10.2=hba1933b_1
• py-opencv==3.4.2=py37hb342d67_1
• ffmpeg==4.0=hcdf2ecd_0
• pip==21.0.1=py37h06a4308_0
• lcms2==2.11=h396b838_0
• intel-openmp==2020.2=254
• libvpx==1.7.0=h439df22_0
• libglu==9.0.0=hf484d3e_1
• matplotlib-base==3.3.4=py37h62a2d02_0
• bzip2==1.0.8=h7b6447c_0
• dbus==1.13.18=hb2f20db_0
• mkl_random==1.1.1=py37h0573a6f_0
• openssl==1.1.1j=h27cfd23_0
• certifi==2020.12.5=py37h06a4308_0
• fontconfig==2.13.1=h6c09931_0
• graphite2==1.3.14=h23475e2_0
• cudatoolkit==10.0.130=0
• ninja==1.10.2=py37hff7bd54_0
• tk==8.6.10=hbc83047_0
• matplotlib==3.3.4=py37h06a4308_0
• libedit==3.1.20191231=h14c3975_1
• cairo==1.14.12=h8948797_3
• freeglut==3.0.0=hf484d3e_5
• sqlite==3.33.0=h62c20be_0
• gst-plugins-base==1.14.0=h8213a91_2
• jpeg==9b=h024ee3a_2

Do I need to install all these dependencies individually or should the above command have worked as it is?
Please help me out as installing them individually is also not working with pip. Further, let me know if I need to use some package manager other than pip for individual installation.

Thanks in advance!

hello,I read read.me carefully ,but i dont find the file i say .

delta_map = kin_chain[:, :, np.newaxis, np.newaxis].repeat(32, axis=-2).repeat(32, axis=-1)
It seems to make the value in W/H the same. is it right? We only need to train the corresponding position of the groundtruth. But we find the position of maximum value of the predicted heatmap to match the groundthuth.

as in

Hello,the final result of your code is a model of OBJ file (which can be imported into 3D Max) or is it just a hand of 3d image displayed in the picture, such as looking at the picture representation:

or obj file like this:

Hello , I have install open3d properly , but when I run the demo, can't show the result, just white board(both demo.py and demo_dl.py), I wonder whether some result error?

Hi, on running demo.py, i am getting error

PS D:\KISHAN FILES\Minimal-Hand-pytorch-main> python demo.py
load model start
Traceback (most recent call last):
File "demo.py", line 16, in
check_point = torch.load('new_check_point/ckp_detnet_83.pth', map_location=device)
File "C:\Users\saisr\AppData\Local\Programs\Python\Python37\lib\site-packages\torch\serialization.py", line 579, in load
with _open_file_like(f, 'rb') as opened_file:
File "C:\Users\saisr\AppData\Local\Programs\Python\Python37\lib\site-packages\torch\serialization.py", line 230, in _open_file_like
return _open_file(name_or_buffer, mode)
File "C:\Users\saisr\AppData\Local\Programs\Python\Python37\lib\site-packages\torch\serialization.py", line 211, in init
super(_open_file, self).init(open(name, mode))
FileNotFoundError: [Errno 2] No such file or directory: 'new_check_point/ckp_detnet_83.pth'

where is this .pth file 4?

it seems the input image is not normalized to [0,1] in the demo.

Hi, thanks for your great work.
I think there is a small bug at

If the batch size is equal to 1, the first dimension will also be squeezed.
I think it should be squeeze(2) instead of squeeze().
Thank you.

class detnet(nn.Module):
        def forward(self, x):
                features = self.resnet50(x)
        
                device = x.device
                pos_tile = get_pose_tile_torch(features.shape[0]).to(device)
                x = torch.cat([features, pos_tile], dim=1)

could you please explain what is the purpose of this pos_tile and why is needed to concat with features?

In HybrIK paper, I see that

input has 3 parameter, but in your code

only 2.

what is the difference between those?

Hi,

Thanks for this great implementation!

I believe you should drop the requirement for manopth==0.0.1 from your environment.yml file, as it does not exists in PyPI. Just leave the requirement to install it from its Github project.

I find only five fingertips are used in these two datasets. Is this a common practice?
This original paper said they use these five joints for global scale and translation alignment, but did not mention that they just use these five joints for evaluation. Why not evaluate all 21 joints after alignment?

hello, do u have some idea about projecting the output's mesh to original image approximatively? Of course assume no camera intrict parameters. I think the accuracy of mesh can be judged more accurately when mesh is projected into the image.just like the paper show:

i am getting this issue on running demo.py. can you help me in resolving this error.

hi, great work for 3D hand ,and I reproduce the shape net but the shape L2 regularizer loss alwary go up, whether it is normal？

when I run demo.py, I got the following pro:
File "demo.py", line 40, in <module> weight=1e-5) File "/storage07/wuwenbin02/Projects/minimal-hand/utils/LM_new.py", line 38, in __init__ _, self.joints = self.mano_layer(self.th_pose, self.th_beta) File "/storage07/wuwenbin02/anaconda3/envs/minimal-hand-torch/lib/python3.7/site-packages/torch/nn/modules/module.py", line 547, in __call__ result = self.forward(*input, **kwargs) File "/storage07/wuwenbin02/anaconda3/envs/minimal-hand-torch/lib/python3.7/site-packages/manopth/manolayer.py", line 183, in forward th_betas.transpose(1, 0)).permute( RuntimeError: size mismatch, m1: [2334 x 10], m2: [48 x 1] at /opt/conda/conda-bld/pytorch_1565272271120/work/aten/src/TH/generic/THTensorMath.cpp:752

i got that error after trying to run demo.py

line 3 :from manopth import manolayer
ImportError: cannot import name 'manolayer'

under folder manopth i can see only one file (rotproj.py)

i have the file in this directory ,but still have this error,how can i solve it ?

hello. I changed the path to an absolute path, so it should be no problem.but another question happen. i dont change this code.i dont know why this is .

I want to know this error how to sovle it ?

I have recently connected the original code to Unity through fast sockets. It looks easier and more powerful than open3d or matplotlib.

https://github.com/vinnik-dmitry07/minimal-hand

Hi, i am not get the third window i.e 3d plot of keypoints of hand. Can you suggest any solution.

Can you tell me that you test on GPU or CPU? THX

based on your experiments log, does analytical IK improves the performance over IKNet?

Hi,
I trained model following the instruction completely. But I got the result is that:

another your model result:

Hello, how should this error be corrected

Traceback (most recent call last):
File "D:/Work/writing/Minimal-Hand-pytorch-main/Minimal-Hand-pytorch-main/demo.py", line 60, in
hand_verts, j3d_recon = mano(pose0, shape.float())
File "D:\software\anaconda\envs\yolov5\lib\site-packages\torch\nn\modules\module.py", line 722, in _call_impl
result = self.forward(*input, **kwargs)
File "D:\Work\writing\Minimal-Hand-pytorch-main\Minimal-Hand-pytorch-main\manopth\manolayer.py", line 196, in forward
root_trans = th_with_zeros(torch.cat([root_rot, root_j], 2))
RuntimeError: All input tensors must be on the same device. Received cuda:0 and cpu

hi, when I check the train code. I found that the kpt order convertion in different dataset(convert to SNAP's order) maybe has some bug. just like STB to SNAP order, the implement is:


I think it will not convert properly. where should like this?

"resnet_helper.py" this file have a missing for a file ,see in the following :

I try to run the demo.py, but get the following errors:

load model start
load model finished
start opencv
opencv finished
Traceback (most recent call last):
  File "demo.py", line 54, in <module>
    hand_verts, j3d_recon = mano(pose0, shape.float())
  File "/home/zhy/anaconda3/envs/minimal-hand-torch/lib/python3.7/site-packages/torch/nn/modules/module.py", line 547, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/zhy/anaconda3/envs/minimal-hand-torch/lib/python3.7/site-packages/manopth/manolayer.py", line 188, in forward
    self.th_posedirs, th_pose_map.transpose(0, 1)).permute(2, 0, 1)
RuntimeError: Expected object of backend CPU but got backend CUDA for argument #2 'mat2'

How to fix this bug?

Hi, there
I've been trying to run the code but failed, this is the error I get.
el start
load model finished
start opencv
opencv finished
Traceback (most recent call last):
File "C:/Users/1/PycharmProjects/Minimal-Hand-pytorch-main/demo.py", line 55, in
hand_verts, j3d_recon = mano(pose0, shape.float())
File "E:\Users\1\anaconda3\envs\yolo\lib\site-packages\torch\nn\modules\module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "C:\Users\1\PycharmProjects\Minimal-Hand-pytorch-main\manopth\manolayer.py", line 187, in forward
th_v_posed = th_v_shaped + torch.matmul(
RuntimeError: Tensor for 'out' is on CPU, Tensor for argument #1 'self' is on CPU, but expected them to be on GPU (while checking arguments for addmm)
[ WARN:0] global C:\Users\appveyor\AppData\Local\Temp\1\pip-req-build-n4ekh6o5\opencv\modules\videoio\src\cap_msmf.cpp (435) `anonymous-namespace'::SourceReaderCB::~SourceReaderCB terminating async callback

Any ideas as to how to fix this? Thanks in advance

there is a error,how to solve ?

RuntimeError: G:/Minimal-Hand-pytorch/my_results/checkpoints/ckp_siknet_synth_41.pth.tar is a zip archive (did you mean to use torch.jit.load()?)

Sorry, this error occurred when I was running demo.py. Has this file been deleted? Thank you ！

Hello, have this files been deleted? thanks!

hello , there are three questions
1.your data is STB datasets I cant open the website,can you send my email ([email protected])？ thank you!
2.another file data “dexter+object” is different of data structure see the follow picture.

in my unzip file "dexter+object",there is missing file "bbox_dexter+object.csv" and "Do_pred_2d.npy"。
3.in code demo.py -"check_point = torch.load('new_check_point/ckp_detnet_83.pth', map_location=device)"
run demo.py ，then result：Traceback (most recent call last):
File "G:/Minimal-Hand-pytorch-main/demo.py", line 16, in
check_point = torch.load('new_check_point/ckp_detnet_83.pth', map_location=device)
File "G:\software-setup\python 3.6.7-setup\lib\site-packages\torch\serialization.py", line 366, in load
load model start
f = open(f, 'rb')
FileNotFoundError: [Errno 2] No such file or directory: 'new_check_point/ckp_detnet_83.pth'
"new_check_point/ckp_detnet_83.pth" the file i don't find it.
please answer this .thank you !

Hello, I encountered this problem when running the demo. I have upgraded open3d to 0.9, but I still can't solve it. Do you have a better way？