I meant to use this project to show I can work with images, but the task turned out to be too easy (and I don't have much more data to improve it further - only 127 train and 30 validation images were used).
Tech-wise I only just used most of yolov7 code as is.
Nevertheless, I uploaded the simple data converter from labelstudio's json format to yolo format. The sample json format can be found in data/raw
Then the results from model prediction can be found in the 2 videos below.
