This repository contains some DataSet Generation and Evaluation Tools and an adapted Pixel2Mesh implementation for the following paper

Pixel2Mesh: Generating 3D Mesh Models from Single RGB Images (ECCV2018)

Check Pixel2Mesh Repository for more information on how to set up Pixel2Mesh.

1. Datset Downloader

• Initial Step: Download your desired .obj files from ShapeNet or Google 3D Warehouse

2. Dataset Tools

• Second Step: Prepare your DataSet for Pixel2Mesh or other Neural Networks.
  • Includes the Renderer to generate png's from different viewpoints.
  • Includes Occlusion (cropping holes in png's)
  • Generating Training and Testing Split

3. Pixel2Mesh

• Run your desired Neural Network (in our case Pixel2Mesh) with different variants:
  • Pixel2Mesh with 2D (standard implementation)
  • Pixel2Mesh with 0.5D (only depth images)
  • Pixel2Mesh with 2.5D (rgbd images)

4. Evalution Tools

• Some Tools for plotting losses
• Losses per viewpoint analysis

• Python2.7+ with Numpy and scikit-image

• Tensorflow (version 1.0+)

• TFLearn

• Code has been tested with Python 2.7, TensorFlow 1.3.0, TFLearn 0.3.2, CUDA 8.0 on Ubuntu 14.04.

• Python3
• BeautifulSoup, joblib, pandas, requests, numpy

• Python3
• Working blender (check Renderer Readme)
• (Meshlab)