commaVQ - GPT2M

A GPT2M model trained on a larger version of the commaVQ dataset.

This model is able to generate driving video unconditionally.

Below is an example of 5 seconds of imagined video using GPT2M.

<video title="imagined" controls> <source src="https://github.com/commaai/commavq/assets/29985433/f6f7699b-b6cb-4f9c-80c9-8e00d75fbfae" type="video/mp4"> </video>