Nice work - can you share how large your training datasets are in terms of number of images? And did you train your models from scratch or fine-tune them from an existing model?