Pset-3d

Installation

In order to set up the environment in google Colab, instead of creating conda environment shown in the original github page, we only need to install several dependencies.

Change the Python version

Directly installing dependencies using the python version in Colab may cause some errors and issues. Therefore, hereby provide a way of environment setup solving this:

Firstly, you can degrade the python version of Colab using (you may find changing python version useful)

!sudo apt-get install python3.7
!sudo update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.7 1

The second line will prompt you to choose a python version among all versions you have. Then run following to make it work

!sudo update-alternatives --config python3

Note: you can always check the python version using !python --version.

Since the python version is now changed, run the following to keep the version of pip and distutils compatible with python

!sudo apt install python3-pip
!sudo apt install python3.7-distutils

The following can be run to install several useful packages

!pip install numpy cython pandas plyfile scikit-image

Compile the extension modules

Next, we can compile the extension models via

python setup.py build_ext --inplace

Note: issues may come up when running this command and you may solve with the following:

1. You may need to modify setup.py a little bit if meeting compiling errors following compiling issues.
2. You may need to run the following before compiling (see resource that may be useful setup script exited with error: command 'x86_64-linux-gnu-gcc' failed with exit status 1)

!apt-get install python3.7-dev
!sudo apt-get install build-essential python3-dev libssl-dev libffi-dev libxml2 libxml2-dev libxslt1-dev zlib1g-dev

Install PyTorch and other dependencies

After successfully compiling the modules, you can install PyTorch dependencies and other necessary packages. Run the following to install PyTorch

!pip install torch==1.9.0+cu111 torchvision==0.10.0+cu111 -f https://download.pytorch.org/whl/torch_stable.html

Then install torch_scatter using

!pip install torch-scatter==2.0.9 -f https://pytorch-geometric.com/whl/torch-1.9.0+cu111.html

Note: the torch version and cuda version for torch and torch_scatter should keep same.

Finally, install other dependencies with

!pip install tensorboardx==1.4
!pip install matplotlib trimesh pyyaml tqdm

Demo

You can run a simple demo to see if the environment is successfully set up. This part directly follows demo in the original github page for Convolutional Occupancy Networks. You can simply run the demo to reconstruct a synthetic indoor scene.

First, run the script to get the demo data:

bash scripts/download_demo_data.sh

Next, generate a mesh from synthetic room dataset:

python generate.py configs/pointcloud/demo_syn_room.yaml

The generated mesh data can be found in out/demo_syn_room. The mesh file .off can be simply visualized using MeshLab.

Data Preparation

In this project, we are going to use the data of Stanford Bunny, which can be downloaded from bunny_data. Data preparation follows the [Occupancy Networks] (https://github.com/autonomousvision/occupancy_networks#preprocessed-data). After git clone the project of Occupancy Networks, enter the project directory and run the following to sample the data:

!mkdir -p prepared_data/bunny/pointcloud \
      prepared_data/bunny/points \
      prepared_data/bunny/mesh

%cd scripts/

# sample the mesh to obtain pointcloud
!python sample_mesh.py ../bunny_data/mesh \
      --pointcloud_folder ../prepared_data/bunny/pointcloud \
      --points_folder ../prepared_data/bunny/points \
      --mesh_folder ../prepared_data/bunny/mesh \
      --packbits --float16 \

This will create the dataset sampled on mesh of Stanford Bunny in 'prepared_data': the output (scaled) mesh, query points and output pointcloud data. You may want to use more input argument options (--points_uniform_ratio, --points_sigma) to apply different sampling strategy. Later you may need to consider adjust these to achieve better performance. Since we only use one object, i.e. a single Stanford Bunny, for training and testing, the splitt files for train/val/test data would be the same and can be created either by modifying 'scripts/create_split.py' or using the following:

# prepare overfit dataset, i.e. one object for train, val and test
import os
import random


in_folder = "../prepared_data"

all_samples = [name for name in os.listdir(in_folder)
               if os.path.isdir(os.path.join(in_folder, name))]


train_set = val_set = test_set = all_samples[:3]

with open(os.path.join(in_folder, 'train.lst'), 'w') as f:
    f.write('\n'.join(train_set))

with open(os.path.join(in_folder, 'val.lst'), 'w') as f:
    f.write('\n'.join(val_set))

with open(os.path.join(in_folder, 'test.lst'), 'w') as f:
    f.write('\n'.join(test_set))

Now you should have all the split files end with .lst inside the prepared_data folder, which is ready for training and testing.

Training

Test and Evaluation