search

xiexh20 / behave-dataset

Code to access BEHAVE dataset, CVPR'22
https://virtualhumans.mpi-inf.mpg.de/behave/
Other
147 stars 7 forks source link

I can't see the mesh in the video #1

Closed Daixinxin-cc closed 2 years ago

Daixinxin-cc commented 2 years ago

Hi,

I have run your code, but the result in the video is like this, I can't see the mesh of kinect1 and kinect2. what's the problem? 微信图片_20220508104755

xiexh20 commented 2 years ago

hi, which sequence did you use? can you verify the SMPL registration is there inside the folder SEQ_NAME/frame_time/person/?

Daixinxin-cc commented 2 years ago

hi, which sequence did you use? can you verify the SMPL registration is there inside the folder SEQ_NAME/frame_time/person/?

I use Date01_Sub01_backpack_back, the content of person folder 1652107254(1) the content of fit02 folder d5f17160d4c6d3b7a266410c39fc5a2

I think the person_fit.ply is the SMPL registration???

xiexh20 commented 2 years ago

Hi, I checked this sequence, it does not have annotation for frame t0003.000. And the visualization code won't load this frame either as it is not there. Where did you get this frame?

Daixinxin-cc commented 2 years ago

Hi, I checked this sequence, it does not have annotation for frame t0003.000. And the visualization code won't load this frame either as it is not there. Where did you get this frame? i can't see the mesh of each frame of each sequence, not one frame and one sequence i download the behave data on your websit

xiexh20 commented 2 years ago

I am able to render the annotations for the sequences you mentioned. Did you follow my instructions exactly? and what did you get after running this demo command? python behave_demo.py -s BEHAVE_PATH/sequences/Date04_Sub05_boxlong -v YOUR_VISUALIZE_PATH -vc

Daixinxin-cc commented 2 years ago

I am able to render the annotations for the sequences you mentioned. Did you follow my instructions exactly? and what did you get after running this demo command? python behave_demo.py -s BEHAVE_PATH/sequences/Date04_Sub05_boxlong -v YOUR_VISUALIZE_PATH -vc

like this https://user-images.githubusercontent.com/57379418/167563126-f5f16a78-9aa4-4acb-91f9-ef58cdb4466b.mp4

xiexh20 commented 2 years ago

I am not able to reproduce your results. Please provide detailed instructions to reproduce your issue. Or to further debug, you could check if the rendered image here https://github.com/xiexh20/behave-dataset/blob/main/behave_demo.py#L70 is fully white or not

mnauf commented 2 years ago

I am having the same issue. @xiexh20 to reproduce this issue, please try this out on a new Linux machine. You might have a few required dependencies already installed on your existing machine.

P.S. @Daixinxin-cc were you able to solve it?

mnauf commented 2 years ago

@bharat-b7 Will you also be able to look into this, please

Daixinxin-cc commented 2 years ago

I am having the same issue. @xiexh20 to reproduce this issue, please try this out on a new Linux machine. You might have a few required dependencies already installed on your existing machine.

P.S. @Daixinxin-cc were you able to solve it?

No. so you solve this on a new linux machine?

mnauf commented 2 years ago

I dumped the result of line 70 using cv2.imwrite(). This is what it shows. This is an entire black image. @Daixinxin-cc Is this the same for you? lala

xiexh20 commented 2 years ago

@mnauf it is normal that the rendered image is fully black, because the rendered image is in the range of 0-1. it is converted to 0-255 before saving otherwise the image will be black. which OpenCV version are you using? if you are using a conda env, can you share your env configuration? (i.e. the output of conda env export > environment.yml)

mnauf commented 2 years ago

cv2 version is 4.5.5.64. Sending yml file in a while

mnauf commented 2 years ago

@xiexh20 Please find the file here https://drive.google.com/file/d/1uSEaj7BSCa1QjPklm-utbwglHDNDQCJT/view?usp=sharing

xiexh20 commented 2 years ago

Hi, it is due to a pytorch3d bug, see issue. The latest pytorch3d (0.6.2) has a bug for the PerspectiveCamera projection, I pushed a quick fix for this. But I suggest you use pytorch3d<=0.5 until this bug is fixed.

mnauf commented 2 years ago

@xiexh20 Worked for me. Here are the results. Thanks a ton! I think we can now mark the issue as resolved.

supriya305 commented 2 years ago

I am getting the same error as @Daixinxin-cc. I have pytorch3d version 0.3. I tried it for following sequences, Date01_Sub01_backpack_back, Date03_Sub03_stool_lift, Date03_Sub04_backpack_hand, Date03_Sub05_basketball etc. For all, I was not able to see the mesh in both kinect1 and kincet2. image

I am using conda env. This is my environment.yml file. https://drive.google.com/file/d/1dwj81iFLQdIVc2m77KVS3WV0N38M38Zh/view?usp=sharing @xiexh20 Can you please look into this?