I built the environment as you specified and ran the command

!python run_inversion.py --name author_released --checkpoint_dir pretrained

Then I got the following error

Traceback (most recent call last):.
File "run_inversion.py", line 12, in <module
from lib.mesh_inversion import MeshInversion
File "/content/drive/MyDrive/mesh-inversion/lib/mesh_inversion.py", line 14, in
from lib.rendering.renderer import Renderer
File "/content/drive/MyDrive/mesh-inversion/lib/rendering/renderer.py", line 1, in
from kaolin.graphics.dib_renderer.rasterizer import linear_rasterizer
ModuleNotFoundError: No module named 'kaolin.graphics'.

I don't know what is causing this, do you know how to solve it?
I am using googlecolab.

Also, which file should I put the photos and other files that I want to reconstruct in 3D? It would be great if you could tell me all the file structure as well.

Thank you in advance for your help.

Hello, thank you very much for your excellent work, but when I installing kaolin, it mentioned PyTorch >=1.2, <1.5 ，but it is required in your env mentioned PyTorch=1.11 and cuda=11.3. This seems a bit contradictory. Besides, kaolin now supports 0.9 version. I wonder what version of kaolin you installed? Looking forward to your reply, thank you very much!

Thank you for great codes.
But I faced with memory error
My PC has GTX 1050ti (2Gb)
Can you answer Any idea to deal with this problem?

hi, I'm trying to reproduce results and also play around with the model, but I wasn't able to get it to run

the run_inversion.py expects lib/external/ to exist, but it does not, the .gitignore actually ignores it completely. I tried cloning the ChamferDistancePytorch repository but it is missing distChamfer_downsample, could you please share your version so that we could get the model working

Thanks in advanced! and thanks for sharing the pretrained model

Hello, i'm trying to run your work but i'm getting an error when i launch the run_inversion.py script.

Setup

I'm working on a system with:

• 64GB RAM
• 2 RTX 2080ti (12GB VRAM each)
• ubuntu 20.04

Installation procedure i've followed

Kaolin Setup

1. Download kaolin (commit e7e5131)
2. Create environment with: "conda create --name mesh-inversion python=3.7" and then activate it via "conda activate mesh-inversion"
3. Build kaolin following the steps defined in their git page:
   • conda install pytorch==1.11.0 torchvision==0.12.0 cudatoolkit==11.3.1 -c python
   • cd kaolin/
   • python setup.py build_ext --inplace
   • python setup.py install
   • pip install --no-dependencies nuscenes-devkit opencv-python-headless scikit-learn joblib pyquaternion cachetools
   • [test kaolin] => python -c "import kaolin; print(kaolin.__version__)" which prints 0.1.0 after installing the missing lib 'threadpoolctl' via pip

Mesh-Inversion setup

1. Install some requisite packages for mesh_inversion:
   • pip install absl-py packaging tensorboard
2. In run_inversion.py script i've added a print in the for loop just to show the current loop status print(f"Image: {idx}")
   (I get the same error even without this print)

Code Execution

Now i run the mesh_inversion.py script:
python mesh_inversion.py --name author_released --checkpoint_dir pretrained

Error

It takes about 20-30 seconds for the error to appear after the "Image 0" print (which is done in row 38 of run_inversion.py right after the code:
idx = data['idx'][0].item())

Successive Attempts

• I've tried using "--num_workers 0" and "--num_workers 1" -> i still get the same error.
• I've tried using "export CUDA_VISIBLE_DEVICES=0,1" -> i still get the same error.
• I've tried using"--gpu_ids 0,1" with and without CUDA_VISIBLE_DEVICES=0,1 -> i still get the same error.
• I've tried using different cuda versions -> i still get the same error.
• I've also tried with python 3.6 -> i still get the same error.

Do you have any idea of what is going wrong?

Thanks in advance for your support!

Hello, thank you very much for the excellent work, I have one more question, in fact from the visualization of the paper, your method outperforms the existing methods very well, but why does IOU seem to be similar to sota method in quantitative evaluation, and why don't you try training with ground truth mask? Looking forward to hearing from you:)

Solving environment: failed

ResolvePackageNotFound:

• giflib==5.2.1=h7b6447c_0
• portalocker==2.4.0=py37h89c1867_0
• importlib-metadata==4.11.3=py37h89c1867_1
• libwebp-base==1.2.2=h7f8727e_0
• gnutls==3.6.15=he1e5248_0
• readline==8.1.2=h7f8727e_1
• future==0.18.2=py37h89c1867_5
• python==3.7.13=h12debd9_0
• tk==8.6.11=h1ccaba5_0
• setuptools==61.2.0=py37h06a4308_0
• brotlipy==0.7.0=py37h27cfd23_1003
• yaml==0.2.5=h516909a_0
• zstd==1.4.9=haebb681_0
• x264==1!161.3030=h7f98852_1
• nvidiacub==1.10.0=0
• openh264==2.1.1=h4ff587b_0
• libgcc-ng==9.3.0=h5101ec6_17
• intel-openmp==2021.4.0=h06a4308_3561
• _openmp_mutex==4.5=1_gnu
• jpeg==9d=h7f8727e_0
• cryptography==36.0.0=py37h9ce1e76_0
• freetype==2.11.0=h70c0345_0
• gmp==6.2.1=h2531618_2
• mkl_random==1.2.2=py37h51133e4_0
• libstdcxx-ng==9.3.0=hd4cf53a_17
• numpy-base==1.21.5=py37hf524024_1
• certifi==2022.5.18.1=py37h89c1867_0
• pysocks==1.7.1=py37_1
• libprotobuf==3.15.8=h780b84a_0
• libffi==3.3=he6710b0_2
• c-ares==1.18.1=h7f8727e_0
• libidn2==2.3.2=h7f8727e_0
• bzip2==1.0.8=h7b6447c_0
• libgomp==9.3.0=h5101ec6_17
• xz==5.2.5=h7b6447c_0
• ncurses==6.3=h7f8727e_2
• ca-certificates==2022.5.18.1=ha878542_0
• sqlite==3.38.2=hc218d9a_0
• libunistring==0.9.10=h27cfd23_0
• zlib==1.2.11=h7f8727e_4
• nettle==3.7.3=hbbd107a_1
• torchaudio==0.11.0=py37_cu113
• ld_impl_linux-64==2.35.1=h7274673_9
• torchvision==0.12.0=py37_cu113
• libiconv==1.15=h63c8f33_5
• lcms2==2.12=h3be6417_0
• openssl==1.1.1o=h7f8727e_0
• pytorch3d==0.6.2=py37_cu113_pyt1110
• mkl_fft==1.3.1=py37hd3c417c_0
• ffmpeg==4.3.2=hca11adc_0
• cffi==1.15.0=py37hd667e15_1
• lame==3.100=h7b6447c_0
• mkl==2021.4.0=h06a4308_640
• pip==21.2.2=py37h06a4308_0
• libtasn1==4.16.0=h27cfd23_0
• libtiff==4.2.0=h85742a9_0
• mkl-service==2.4.0=py37h7f8727e_0
• libwebp==1.2.2=h55f646e_0
• cudatoolkit==11.3.1=h2bc3f7f_2
• pytorch==1.11.0=py3.7_cuda11.3_cudnn8.2.0_0
• lz4-c==1.9.3=h295c915_1
• libpng==1.6.37=hbc83047_0
• libuv==1.40.0=h7b6447c_0

EnvironmentNameNotFound: Could not find conda environment: mesh_inv
You can list all discoverable environments with conda info --envs.