Collector

This is a small web app to collect training data by presenting the user some input data, and waiting for them to provide the output. It only supports a single type of each, right now, but I have vague plans to make it flexible.

I'm trying to use this to generate training data which is suitable for my very simple control model in rl-sandbox. I want a robot to move towards the red box, but when I manually capture training data from the actual pixels (via my eyeballs) using a simulator, it barely works at all, because at runtime the pixels first pass through a vision model. This program is intended to present me with the same input data that the control model gets, so I can provide examples based only on that.

License

MIT