Does anyone know if it is possible to use ultralytics to import yolo and later add some extra layers?
Transfer learning
Hi @RicardoGuadalupeGomezMartnez
The model has to be trained using luxonis/depthai-ml-trainingblob/master/colab-notebooks/YoloV8_training.ipynb, so I think if you started with ultralytics model, then set it as a checkpoint in the collab, it should work.
Thanks
Jaka
jakaskerl Yes, I entered it as you mention, but my goal is to add layers to the base architecture and subsequently transfer the weights. I understand that it is done from the yolov8.yaml file, but I don't know how the layers are added as well as the syntax.
Hi @RicardoGuadalupeGomezMartnez
Right, misread the layer part.
cc @Matija any ideas of if this is possible/how to achieve it?
Thanks,
Jaka
What layers would you like to add RicardoGuadalupeGomezMartnez ?
It depends where you will be adding the layers. We cut off some post-processing in the heads output, so that we can decode this correctly on device. Even if you add base layers, this should be possible, but the "semantic meaning" of the outputs in modified model should still be the same.