3D mesh is not generated about meshtransformer

[tqtrunghnvn]

Thank you so much! The problem is solved.

[kevinlin311tw]

Could you send your test sample to me? I can try running it.

[tqtrunghnvn]

I simply downloaded an image from the internet. You can try with this following one:
https://keyassets.timeincuk.net/inspirewp/live/wp-content/uploads/sites/64/2018/12/champ-630x473.jpg

[kevinlin311tw]

Three more questions:

1. Are you able to generate our 3D mesh examples? This is to make sure you have install the code correctly.
2. Before feeding your golf image, do you crop the image with a size of 224x224? We need 224x224 image size.
3. Could you please share any error messages you got here?

[tqtrunghnvn]

Hi,
I was not able to generate the 3D mesh for your examples.
I got:

Could you please try to run another sample?

[kevinlin311tw]

You may find it here.

In your attached results, I see you have generated the attention visualization with the output vertices. This indicates 3D mesh is generated. The issue you got is probably a rendering issue. I would suggest to double check the installation, making sure OpenDR is successfully installed.

[tqtrunghnvn]

Hmm. Thank you for your effort!

I have double-checked opendr installation, and is successfully installed and could be imported without any problems.
Could you tell me where is the code for generating 3D mesh (NOT attention visualization)?

[kevinlin311tw]

I guess you are referring to rendering 3D mesh, right? You may check it here:

MeshTransformer/metro/utils/renderer.py

Line 405 in 44760db

rend_img = renderer.render(vertices_full, camera_t=camera_t,

Basically it takes mesh vertices, camera parameters, input image as inputs. It will render 3D pink mesh on top of the image.

[tqtrunghnvn]

Is it mandatory to install the dependencies with the following versions?

• Python 3.7
• Pytorch 1.4
• torchvision 0.5.0
• cuda 10.1

[kevinlin311tw]

We verified that our code works with the dependencies. I would suggest to literally follow our installation instruction, and make sure you can reproduce our demo examples.

[zws2lll]

@tqtrunghnvn hi, tqtrunghnvn, could you tell me how to solve this problem? I run the demo on gpu 3090.

[tqtrunghnvn]

Hi @zws2lll , you should check your opendr installation.

[zws2lll]

@tqtrunghnvn I use "pip install opendr", and it's 0.78. There's no error. How do you check the opendr?

Successfully built opendr
Installing collected packages: opendr
Attempting uninstall: opendr
Found existing installation: opendr 0.76
Uninstalling opendr-0.76:
Successfully uninstalled opendr-0.76
Successfully installed opendr-0.78

[tqtrunghnvn]

You could try:
https://gitlab.eecs.umich.edu/ngv-python-modules/opendr

[zws2lll]

I clone it and "pip install .". 0.76 is installed. The output contains no render hand

./metro/tools/end2end_inference_handmesh.py --resume_checkpoint ./models/metro_release/metro_hand_state_dict.bin --image_file_or_path ./samples/hand/

[zws2lll]

I clone it and "pip install .". 0.76 is installed. The output contains no render hand

./metro/tools/end2end_inference_handmesh.py --resume_checkpoint ./models/metro_release/metro_hand_state_dict.bin --image_file_or_path ./samples/hand/

@kevinlin311tw hi, kevinlin311tw. I also use your docker image. Same result with no render hand. Any Suggestion？

[tqtrunghnvn]

@zws2lll I also used opendr0.76. Sorry I have no idea now

[zws2lll]

Thank you anyway.

[fmx789]

Thank you anyway.

@zws2lll Hi, have you solved the problem? Such a circumstance perplexes me as well.