This workflow is intended to provide a head start for anyone needing to annotate image sets for yolo custom model training. It processes your image files using Florence2 for initial bbox detection and then SAM2 for finer bbox detection. It is intended to save files in a "/yolo/" sub-directory relative to the image base directory specified. It saves padded 640x640 images (default yolo dimensions) in a "/images" sub-directory and the labels which include the bbox dimensions in yolo format in the "/labels" sub-directory.If you build on this, please hit me up!- BillyB

YOLO Dataset Auto-Annotate Workflow

This version adds the ability to select which bbox to use in the situation that Florence2Run returns more than one. Typically, I have found that the higher the bbox index, the smaller the area and the more tightly matching the description. Thus, I wanted a way to be able to pick the tightest match, i.e. the bbox with the smallest area. After learning about For Loops, I've added functionality to allow you, the user, to specify that.

Files with SHA256: b8a064cedacbf4978acb9ae9eca78e662ea0830c71269f628d16a076223c5529