/Pytorch3DGenRenderer

A compact tool for rendering meshes built upon Pytorch3D library

Primary LanguagePython

3D Mesh Renderer

This project is designed to render 3D meshes from various angles, which is particularly useful for creating datasets for 3D deep learning tasks.

Table of Contents

Installation

  1. Install Pytorch3D: Before proceeding, ensure you have Pytorch3D installed. It's recommended to follow the official instructions.

  2. Install the Project:

    pip install -e .

Data and Configuration Preparation

  1. Generate Configuration Files: For each 3D model you wish to render, generate a YAML configuration file and store them in a specific folder. An example of how to generate these configurations is provided in config_generation_example.py.

    The configuration supports the following parameters:

    • image_size: Size of the output renders in pixels.
    • num_samples: Number of output rendered images.
    • randomize: Parameter for randomizing angles of view. Important for some sampling methods.
    • obj_filename: Path to the .obj file to render.
    • sampling_method: Method name to sample points of view for rendering.
    • seed: Fixes the distribution of angles if randomizing is applied.
    • save_dir: Directory to save output renders and other files.
    • save_gif: Boolean parameter to save a GIF consisting of output renders.
    • save_vis: Boolean parameter to save an HTML file visualizing the distribution of camera angles on a sphere.

    Here's a configuration example:

    image_size: 1024
num_samples: 20
randomize: true
obj_filename: data/cow/cow.obj
sampling_method: fibonacci_sphere
seed: 42
save_dir: data/cow
save_gif: false
save_vis: false

  2. Create a Rendering Configuration: This configuration dictates the rendering process. It requires the following parameters:

    • params_dir_path: Path to the directory containing the above-described configurations.
    • num_gpus: Number of GPUs to utilize for rendering. A value of -1 means to use all available GPUs.

    Here's an example:

    params_dir_path: params
gpus_num: -1

Examples of renders of aircraft are provided below:

Usage

Once you've prepared all the necessary configurations, you can initiate the rendering process with the following command:

render --config [your_render_config.yaml]