This is an excellent model! I'm curious—how did you source your dataset? Did you rely on annotation by large-scale models, specialized models, or was it done manually? And where did you download the images?