Collada to Pix3d Format

This can be used to convert the model in .dae to the pix3d format so that you can train your 3d Reconstrution Model.

Requirements

• Blender
  • Numpy
  • Opencv
  • pycocotools
• binvox
• scipy

Installation

You need to install the packages mentioned above under the blender to the python which comes bundled with blender.
The Packages can be installed to python bundled with the following steps

1. Clone the repo

git clone https://github.com/AnuragSahu/pix3d_from_dae.git
cd pix3d_from_dae/

2. Setup Blender

Method 1 (Long) Download and Install

a. Get Blender

wget https://ftp.nluug.nl/pub/graphics/blender/release/Blender2.83/blender-2.83.1-linux64.tar.xz
tar -xf blender-2.83.1-linux64.tar.xz 
mv blender-2.83.1-linux64.tar.xz ./blender
cd blender

b. Install packages

cd 2.83/python/bin/
./python3.7m -m ensurepip
./python3.7m -m pip install opencv-python
./python3.7m -m pip install numpy
./python3.7m -m pip install pycocotools
cd ../../../../

Method 2 (Short) Download my Blender

a. Get the blender from Google Drive

git clone https://github.com/AnuragSahu/download_to_Server_from_Google_drive.git
cd download_to_Server_from_Google_drive/
python download_gdrive.py 1Yt-FS-_ppg0-j9CZgxdIGIjaptazKIhN ../blender.zip
cd ../
rm -rf download_to_Server_from_Google_drive

b. Unzip Blender

cd ..
unzip blender.zip

3. Get the binvox and scipy, binvox is already given with the github repo

pip install scipy

Get the datasets

1. Setup, You should place you models in ./pix3d_train/models with the following file structure

models
    |
    |----> Object1_Name
                |
                |----> model.dae
    |
    |----> Object2_Name
                |
                |----> model.dae

*note that the name model.dae should not be changed
2. Run the code

cd ./blender/
./blender --background -P ../pix3d_dataset/src/generate_pix3d.py

Note : To get the locations file I have created a YouTube Video Link