Jetbot Object Detection Model Issue - Cannot load

Good day, I am trying to troubleshoot a JetBot issue, this is a Waveshare jetbot kit, installed using the latest Nvidia Image and then WaveShare’s installer.

When I attempt to run the object detection script

from jetbot import ObjectDetector
model = ObjectDetector('ssd_mobilenet_v2_coco.engine')

I get the following output.

AttributeError                            Traceback (most recent call last)
<ipython-input-1-78c468b59bd4> in <module>
      1 from jetbot import ObjectDetector
----> 3 model = ObjectDetector('ssd_mobilenet_v2_coco.engine')

/usr/local/lib/python3.6/dist-packages/jetbot-0.3.0-py3.6.egg/jetbot/ in __init__(self, engine_path, preprocess_fn)
     26         load_plugins()
     27         self.trt_model = TRTModel(engine_path, input_names=[TRT_INPUT_NAME],
---> 28                                   output_names=[TRT_OUTPUT_NAME, TRT_OUTPUT_NAME + '_1'])
     29         self.preprocess_fn = preprocess_fn

/usr/local/lib/python3.6/dist-packages/jetbot-0.3.0-py3.6.egg/jetbot/ in __init__(self, engine_path, input_names, output_names, final_shapes)
     57         with open(engine_path, 'rb') as f:
     58             self.engine = self.runtime.deserialize_cuda_engine(
---> 59         self.context = self.engine.create_execution_context()
     61         if input_names is None:

AttributeError: 'NoneType' object has no attribute 'create_execution_context'

Some have suggested this is due to TensorRT version. The version in this jetpack is 6.0 - some say I need 5. I have tried re-compiling TRT and a few other things to no avail.

Any advice appreciated, thanks.

I have the same issue with a Jetbot built directly from the wiki. Everything updated, latest version.

The examples all show TensorRT 5, but a PIP3 list whows TensorRT 6, so I am thinking that running this latest JetPack is part of the problem. I may try downgrading it, I may also have to take the WaveShare bits apart to and ensure they are not updating it.

I was able to get it working. Thanks to this thread…

The JetBot Version 4 SD Card Image ships with TensorRT 6, which, for a reason I cannot figure out simply doesn’t want to work. By downgrading th SD Card Image to JetBot Version 3, and then applying my kit makers specific modification (In my case WaveShare) I was able to get it working.

Hi jucohen, Thanks for the details you provided. In addition to what you did, it’s worth retrying with a new Jetbot V 0.4 SD image and trying the example again before updating anything. In my case, I got past the model load issue. The object following doesn’t work too well but that may be a separate issue, perhaps due to the model having been trained with the different camera than the LI IMX219 that I have.

The latest ssd_mobilenet_v2_coco.engine is here.

v0.4 (latest) ssd_mobilenet_v2_coco.engine