RetinaNet model conversion from Pytorch to Openvino

[Akash]

Hi,

I am following the this https://github.com/xytpai/retinanet  repository training Retinanet model using Pytorch. I have trained retinanet model with Resnet-50 as backbone.

Facing below error while converting the onnx model to Openvino IR.

python3 mo_onnx.py --input_model model_conversion/gloves.onnx --output_dir  model_conversion/IR/ --input_shape [1,3,641,641]

[ ERROR ] Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.front.user_data_repack.UserDataRepack'>): Original placeholders: 'input, loc'. Freezing was requested for ''. --input_shape was provided without --input. Can not deduce which node shape to override..

Configurations are as below.

openvino =2020.2.120

Onnx = 1.5.0

I have uploaded my model to drive

https://drive.google.com/file/d/1uPSjv5pRM_RdskRA3wxCF78oMW_vO0Dd/view?usp=sharing 

Please find the onnx model in above link.

Thanks.

[Munesh_Intel · Accepted solution]

Hi Akash,

RetinaNet using PyTorch is not supported officially by OpenVINO.

Details about supported PyTorch models are available at the

https://docs.openvinotoolkit.org/2020.4/openvino_docs_MO_DG_prepare_model_convert_model_Convert_Model_From_ONNX.html#supported_pytorch_models_via_onnx_conversion

Alternatively, OpenVINO officially supports RetinaNet with Resnet 50, from https://github.com/fizyr/keras-retinanet. The relevant information is available at the following page:

https://docs.openvinotoolkit.org/2020.4/omz_models_public_retinanet_tf_retinanet_tf.html

Regards,

Munesh

[Munesh_Intel]

Hi Akash,

Intel will no longer monitor this thread since we have provided a solution. If you need any additional information from Intel, please submit a new question.

Regards,

Munesh