@aanya_palety

Devlog #2

The model is detecting more fuel now but still not the big one.

Most of my time was spent setting up Jupyter. I had trouble with it because apparently torch doesn’t support Python 3.13 on macOS and then the version of numPy I had doesn’t work on Python 3.10 (which I downgraded to). It took a while and also my sanity but I finally got ultralytics installed and the model is working on Jupyter.

We trained the model with new images and now it can detect robot bumpers. However, its now detecting bumpers where bumpers don’t exist (specifically on that image).

I think that the issue is that the dataset that it was trained on doesn’t have very many images of fuel head on so I’m going to test out a few more similar images and see if it has the same error.

Devlog #1

I have a mostly trained model that is able to recognize fuel (FRC fuel, six inch bright yellow foam balls). I’m working on training it on another almost 1000 images of fuel.

Right now, its having a bit of trouble with one specific image (attached below) so I’m going to start testing similar images and see what the pattern is and why its so unconfident in identifying the fuel.

Hopefully, increasing the dataset will help the confidence and help it properly identify fuel.

It was originally trained on google collab but I’m trying to move it over to JupyterLab (can’t track hours on collab) so a lot of my logged time is from installing everything I need like ultralytics and moving everything into Jupyter. I have spent more than 20 minutes on this though lol. And also collab is annoying and you have to reinstall everything every time you leave or start a new session or are inactive for a bit.

Also, to give credit where credit is due, I’m working with one of my teammates who is helping me until I understand the software. (Also pls dont judge if I said smth wrong or stupid I’m kind of a beginner in this)