Transform the object detection model trained on TensorFlow to iOS CoreML model type used tfcoreml(reference) and coremltools.
- mlmodels_IOU0.4_Conf0.6
Core ML models(iOS) transformed from the models that trained on tensorflow(threshold: IOU=0.4; Confidece=0.6
) - MobileNetV1_SSD/MobileNetV2_SSDLite/MobileNetV3_Large_SSDLite/MobileNetV3_Small_SSDLite/MobileDet_SSD_CPU
The transform process from file list
Original Model trained on tensorflow contains:
preprocess
mobilenet_ssd
anchor generate
NMS
predictor
parts.
- check_modelNodes.py
Load thefrozen_inference_graph.pb
,and print the name of nodes in each layer. Then get the input_node_name(mobilenet) and output_node_name(scores/boundingboxes) - frozenToSimplePB.py
Simplify the model(Strip unused subgraphs get simplified frozen graph-ssd_mobilenet.pb
) - simplePB_ToCoreml.py
Main network tranceform(ssd_mobilenet.mlmodel
) - Decoder.py
- NMS.py
- pipelines.py