Simple Trainer

Training on an image

The examples/ script allows you to test the basic forward projection and rasterization of random gaussians and their differentiability on a single training image. This allows you to overfit gaussians on a single view.

Simply run the script with:
python examples/ --height 256 --width 256 --num_points 2000 --save_imgs

to get a result similar to the one below:

Gaussians overfit on a single image

You can also provide a path to your own custom image file using the --img_path flag:

python examples/ --img_path PATH_TO_IMG --save_imgs