Toyota Design Challenge 2023

The 2023 Toyota design challenge required competitors to detect whether a hole was covered by a sticker, partially covered, or whether there was no sticker on the hole. To make the problem challenging, holes and stickers were similar sizes and detection methods had to detect the three classes with dynamic camera positioning and varied background color.

Addressing the challenge, we created a labeled dataset of 400+ images, implementing transfer learning with YOLO v8. Semi-supervised learning techniques were applied by using model predictions for training. The model achieved 82% accuracy on stickers, 89% on holes, and 92% on partially covered holes. It demonstrated recognition across dynamic camera positions on red, white, and metallic surfaces.

JestureBot Github