Convert pb to uff

[sheena40920]

Hello, I am new to Jetson Nano. I used the MobileNet SSD V2 model after training and saved the ckpt file, and then converted the converted PB frozen file to UFF file, but the final result showed IndexError: List index out of range, how to solve it?
My Jetpack is 4.2.3 and tensorrt is 5.1.6.1frozen_inference_graph.pb (18.5 MB)

[SunilJB]

Hi @sheena40920
UFF and Caffe Parser have been deprecated from TensorRT 7 onwards, hence request you to try ONNX parser.
Please check the below link for the same.

GitHub - onnx/onnx-tensorrt: ONNX-TensorRT: TensorRT backend for ONNX

Thanks

[sheena40920]

Ok i’ll try it, thank you for your reply

[sheena40920]

Sorry, I would like to ask, if I forget to update tensorrt to version 7.2 in github, do I have to follow the latest version to comply with the jetpack before it can be used?

[sheena40920]

Sorry, I would like to ask, before I tried the conversion of pb to onnx using the following URL GitHub - onnx/tensorflow-onnx: Convert TensorFlow, Keras, Tensorflow.js and Tflite models to ONNX
, I have succeeded, but when I inference, it shows that uint8 is not supported. Is there any way to solve it?model.onnx (17.9 MB)

[SunilJB]

Hi @sheena40920 ,

Please refer to below link:

tensorrt inference error while load onnx model

Hi, TensorRT only supports FP32, FP16, INT32, and INT8: https://docs.nvidia.com/deeplearning/sdk/tensorrt-api/python_api/infer/FoundationalTypes/DataType.html#tensorrt.DataType UINT8 is not supported as mentioned there. Can you try to use int8 or int32 instead of uint8 before converting your model to ONNX? Alternatively, you could try to use the ONNX API to convert the UINT8 nodes to INT8 or INT32 after training/converting to ONNX, but these could potentially create incorrect results if not h…

Thanks

[sheena40920]

Sorry, I would like to ask when I was doing training, I did graph_rewriter {in the ssd_mobilenet_v2_quantized_300x300_coco.config file
quantization {
delay: 48000
weight_bits: 8
activation_bits: 8
}
}In the quantization section of this paragraph, the uint 8 appeared when I converted it into the onnx file. What is the reason?ssd_mobilenet_v2_quantized_300x300_coco.config (4.9 KB)

[spolisetty]

Hi @sheena40920,

Could you please let us know how are you converting to ONNX model, we recommend you to follow GitHub - onnx/tensorflow-onnx: Convert TensorFlow, Keras, Tensorflow.js and Tflite models to ONNX

Thank you.

[sheena40920]

Yes, I have always used this github to convert onnx, but the type of my input is still uint8
model.onnx (17.9 MB)

[spolisetty]

Hi @sheena40920,

Looks like this is known issue. Natively TensorRT doesn’t support UINT8 data type. When we generate ONNX model, observed data type is UINT8, modifying ONNX model may be error prone.

Please refer following, you may find useful.

Unsupported ONNX data type: UINT8 (2)

Problem converting ONNX model to TensorRT Engine for SSD Mobilenet V2

Hi everyone, I have trained SSD Mobilenet v2 model on my dataset. It got successfully converted to ONNX but, during converting ONNX model to TensorRT engine, it throws error due to unsupported datatype UINT8. Is their any work around to generate tensorrt engine? If I have to retrain the model with supported datatype, how to change datatype of model from uint8 to supported one? TensorRT version: 7.1 Thank You

If you still need further assistance we recommend you to post your query on Jetson forum to get better help.

Jetson & Embedded Systems

Thank you.