Convert Pytorch Model to CoreML

pytorch-to-coreml.md

Converting models trained using CVNets to CoreML

For conversion, we assume that you are using MAC OS machine. We follow CoreML tutorial, i.e., first convert the PyTorch model to JIT, and then convert to CoreML model.

Classification networks

We can convert the classification models using the following command

export CONFIG_FILE="LOCATION_OF_CONFIG_FILE"
export MODEL_WEIGHTS="LOCATION_OF_MODEL_WEIGHT_FILE"
cvnets-convert --common.config-file $CONFIG_FILE --common.results-loc coreml_models_cls --model.classification.pretrained $MODEL_WEIGHTS  --conversion.coreml-extn mlmodel

Detection networks

We can convert the detection models trained on MS-COCO (81 classes, including background) using the following command

export CONFIG_FILE="LOCATION_OF_CONFIG_FILE"
export MODEL_WEIGHTS="LOCATION_OF_MODEL_WEIGHT_FILE"
export N_CLASSES="NUMBER_OF_CLASSES"
cvnets-convert --common.config-file $CONFIG_FILE --common.results-loc coreml_models_det --model.detection.pretrained $MODEL_WEIGHTS --conversion.coreml-extn mlmodel --model.detection.n-classes $N_CLASSES

Segmentation networks

We can convert the segmentation models using the following command

export CONFIG_FILE="LOCATION_OF_CONFIG_FILE"
export MODEL_WEIGHTS="LOCATION_OF_MODEL_WEIGHT_FILE"
export N_CLASSES="NUMBER_OF_CLASSES"
cvnets-convert --common.config-file $CONFIG_FILE --common.results-loc coreml_models_res --model.segmentation.pretrained $MODEL_WEIGHTS --conversion.coreml-extn mlmodel --model.segmentation.n-classes $N_CLASSES

Example to convert MobileViTv2 model

We can convert MobileViTv2-1.0 classification model trained on ImageNet-1k dataset using below commands:

export CONFIG_FILE="https://docs-assets.developer.apple.com/ml-research/models/cvnets-v2/classification/mobilevitv2/imagenet1k/256x256/mobilevitv2-1.0.yaml"
export MODEL_WEIGHTS="https://docs-assets.developer.apple.com/ml-research/models/cvnets-v2/classification/mobilevitv2/imagenet1k/256x256/mobilevitv2-1.0.pt"
cvnets-convert --common.config-file $CONFIG_FILE --common.results-loc coreml_models_cls --model.classification.pretrained $MODEL_WEIGHTS  --conversion.coreml-extn mlmodel