Open borderlineinteractive opened 6 years ago
@borderlineinteractive The original YOLO model, that was trained on ImageNet and then subsequently Pascal VOC, we unfortunately do not have available as a Turi Create model.
The YOLO9000 model is also a model that we do not have available either as a Turi Create model or as a Core ML model (https://developer.apple.com/machine-learning/build-run-models/).
Thanks for letting us know that this is something that you would find useful and we can think about it for the future.
tinyYolo (50MB weights) flickers with false object detections where full yolo v2/v3 (200MB weights) is amazing for object detection. It would be really powerful to have the full yolo (or comparable accuracy) in turicreate.
Is the original yolo model, which was trained on imagenet available in the Turi Model format? I know that it is available in CoreML format, but I also do not see a way to convert CoreML back to the internal Turi Model format.
It would also be very interesting to have access to a trained model (either Turi of CoreML format) that uses a richer amount of objects, such as yolo9000. A general model like this could be a starting point for more detailed, multistep image analyses.