fix requirements file (conflict and missing modules)

[MvWouden]

Issue

The requirements file is missing some dependencies which are required to train the model or to preprocess custom data.

Changes

1. Bumped numpy to version 1.19.3 to resolve dependency conflict
2. Added module imageio
3. Added module scikit-image
4. Added module tensorboard
5. Added these changes to the "dependencies" in the readme

Steps to reproduce

Installing code and running custom data preprocessing and model training, e.g.

# Clone the repository 
git clone https://github.com/Totoro97/NeuS.git
cd NeuS

# Check python version
python --version  # should be 3.9.X

# Create virtual env
python3 -m venv venv
source venv/bin/activate  # or venv\Scripts\active on windows

# Install correct PyTorch version
pip3 install torch==1.8.0+cu111 -f https://download.pytorch.org/whl/torch_stable.html  # or another (1.8.0)

# Install requirements from file
pip3 install -r requirements.txt

# First issue with requirements file found here
#
# The conflict is caused by:
#     The user requested numpy==1.19.2
#     trimesh 3.9.8 depends on numpy
#     opencv-python 4.5.2.52 depends on numpy>=1.19.3
#
# Solved by bumping numpy to 1.19.3 (previously 1.19.2) and reinstalling dependencies

# Test run according to preprocess_custom_data at https://github.com/Totoro97/NeuS/tree/main/preprocess_custom_data#option-2-use-colmap
cd preprocess_custom_data/colmap_preprocess
python imgs2poses.py ${data_dir}

# Second issue with requirements file found here
#
# ... ModuleNotFoundError: No module named 'imageio'
#
# Solved by adding imageio to requirements file and reinstalling dependencies

# Test run according to preprocess_custom_data at https://github.com/Totoro97/NeuS/tree/main/preprocess_custom_data#option-2-use-colmap
python imgs2poses.py ${data_dir}

# Third issue with requirements file found here
#
# ... ModuleNotFoundError: No module named 'skimage'
#
# Solved by adding scikit-image to requirements file and reinstalling dependencies

# Test run according to preprocess_custom_data at https://github.com/Totoro97/NeuS/tree/main/preprocess_custom_data#option-2-use-colmap
python imgs2poses.py ${data_dir}  # Success now!

# Run training
cd ../..
python exp_runner.py --mode train --conf ./confs/womask.conf --case <case_name>

# Fourth issue with requirements file found here
#
# ... ModuleNotFoundError: No module named 'tensorboard'
#
# Solved by adding tensorboard to requirements file and reinstalling dependencies

# Run training
cd ../..
python exp_runner.py --mode train --conf ./confs/womask.conf --case <case_name>

[FlorentLM]

Hey there!

FWIW, while these changes worked for me on Linux, they did not on Windows 11 with CUDA 11.1 - Here is the requirements.txt that did it:

Using a python=3.7.16 environment,

--find-links https://download.pytorch.org/whl/torch_stable.html

trimesh==3.9.8
numpy==1.19.2
pyhocon==0.3.57
icecream==2.1.0
opencv_python==4.5.2.52
tqdm==4.50.2
torch==1.8.0+cu111
scipy==1.7.0
PyMCubes==0.1.2
imageio==2.31.1
scikit-image==0.19.3
tensorboard==2.11.2

[v4r4rth]

Can you please suggest a way to run this on Ubuntu 22.04.4 ? CUDA 11.1 is not supported for me so unsure which torch to get...