Things accomplished this week
- Completed Various Tests for how one would train YOLO to have additional classes that it would classify
- Notably, attempts were made to train on top of existing models but accuracy on existing classes suffered greatly.
- It was decided to train YOLO10m from scratch due its high accuracy and relatively quick training time. From scratch just means borrowing relative classes/photos from COCO and then splicing together existing online datasets of things like pencils.
- There is an ongoing project to see how GPT and YOLO can be used to cross-reference each other.
Is your progress on schedule or behind
1. We are currently on schedule though there need to be things figured out regarding how exactly we will be labelling the images as we will likely be continuously attempting to iterate on the base YOLO design
2. We also need to discuss design choices such as how exactly the database will work but hopefully each unit should be fairly independent of others.
What deliverables do you hope to complete in the next week?
1. I want to have more comprehensive testing of the training from scratch approach. Due to failures relating to the Jetson/rasp pi working all training will be happening locally on my computer.
2. Will coordinate with giancarlo on getting more training data and a better benchmarker
3. Will fully research integrating gpt as a cross reference for YOLO labeling.
0 Comments