Error during Openvino inference of Mask R-CNN Inception ResNet V2 1024x1024

[Vishnuj]

Hi,

I tried to run inference of Mask R-CNN Inception ResNet V2 1024x1024 from Tensorflow2 Model Zoo. I am using the following command to convert the tensorflow model.

mo --saved_model_dir "mask_rcnn_inception_resnet_v2_1024x1024_coco17_gpu-8/saved_model" --transformations_config "/root/sharedfolder/dent_damage_det/mask_rcnn_support_api_v2.0.json" --tensorflow_object_detection_api_pipeline_config "mask_rcnn_inception_resnet_v2_1024x1024_coco17_gpu-8/pipeline.config" --output_dir "/root/sharedfolder/dent_damage_det/mrcnn_zoo" --input_shape [1,1024,1024,3]

The model got converted successfully. While running inference on the converted model I am getting this error:

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-68-f922978d5286> in <module>
      7 request = compiled_model.create_infer_request()
      8 input_image = np.zeros((1, 1024, 1024,3))
----> 9 request.infer(inputs={input_layer.any_name: input_img})

/opt/intel/openvino/python/python3.6/openvino/runtime/ie_api.py in infer(self, inputs)
    108         """
    109         return super().infer(
--> 110             {} if inputs is None else normalize_inputs(inputs, get_input_types(self))
    111         )
    112 

RuntimeError: Proposal operation image info input must have positive image height and width.

Code to run inference:

ie_core = Core()
model = ie_core.read_model(model=model_path, weights=model_weights_path)
compiled_model = ie_core.compile_model(model=model, device_name="CPU")
input_layer = next(iter(compiled_model.inputs))
output_layers = list(compiled_model.outputs)

request = compiled_model.create_infer_request()
input_image = np.zeros((1, 1024, 1024,3))
request.infer(inputs={input_layer.any_name: input_img})

[Zulkifli_Intel]

Hello Vishnuj

Thank you for reaching out to us.

The Python code that you shared appears to be incomplete. You can refer to Inference Pipeline for guidance.

For your information, Intel Open Model Zoo has two mask_rcnn models:

• mask_rcnn_inception_resnet_v2_atrous_coco
• mask_rcnn_resnet50_atrous_coco

Both of these models can be run with Object Detection Mask R-CNNs Segmentation C++ Demo.

Sincerely,

Zulkifli 

[Zulkifli_Intel]

Hello Vishnuj,

Thank you for your question. If you need any additional information from Intel, please submit a new question as this thread is no longer being monitored.

Sincerely,

Zulkifli