Click and drag to rotate the camera. WASD to move.

No longer being developed, may come back to it one day to finish it and upgrade it to WebGPU.

Remaining work involves mapping the trained PyTorch model to a shader. Need to figure out if convolution is implemented as cross-correlation due to learned parameters, and how each kernel is oriented in relation to the image. (E.x. row-major or column-major data?)

This scene doesn't use textures because it was made with the intent of applying a neural network trained with the cityscapes dataset as a shader, but Nvidia beat me to it: