Dan Morris

For now, yes, I would literally just edit your clone of YOLOv5 and either comment out the break statement or raise the time limit. You shouldn't have to change anything...

I agree, that sounds like relatively familiar territory for MD, so I don't think we'll find a satisfying answer in the bounding boxes themselves, this is more likely to be...

Good idea! Since this incompatibility appears to be fixed in PyTorch 1.12 and the latest YOLOv5 (without re-building the model at all), the main purpose of doing this export step...

Thanks... that change (from "map_location=device" to ".to(device)") looks non-controversial; I'll confirm that it has no impact on the existing model in the recommended (PT 1.10, old YOLOv5) environment, and make...

The model is run on every image; the folder structure is used just to support the search box.

OK, I saw this on at least one other machine, and it philosophically drives me bonkers that one might get slightly different results depending on the speed of your machine....

For posterity, the relevant change to the forked repo is [here](https://github.com/ecologize/yolov5/commit/a5b7d2b7554049ea71ef664c4a82bd9bfc9843c4).

Sites change, so a tool like this will need constant maintenance; adding one vote to make @davidjkrause's repo the new home for sportsipy. Please join this bandwagon, and if enough...