search

ros-industrial / easy_perception_deployment

A ROS2 package that accelerates the training and deployment of CV models for industries.
Apache License 2.0
53 stars 16 forks source link

Issue in deploying custom ONNX Model #45

Closed Prabhakaran-Gokul closed 2 years ago

Prabhakaran-Gokul commented 2 years ago

Hi all,

I trained my own custom dataset with YOLOv5s pre-trained model. I want to deploy my model in this package so I followed the instructions given in the documentation. Initially my model had an incompatible model input shape hence, I retrained my model with the required input shape of 224.

I exported my model to .onnx format using this guide (https://github.com/ultralytics/yolov5/issues/251) and made sure that the model performs as expected with the inference code given by YOLOv5 on images with input shape 224. I also created a custom_yolov5_classes.txt file under the label_list folder.

However, no message is published to the /processor/output topic. The Showimage window is also not responding. Here is my terminal output: termial_output_edp

I am able to run the given models(FasterRCNN, MaskRCNN) without any issues. Hope to get your advice on this!

cardboardcode commented 2 years ago

@Prabhakaran-Gokul Thank you for posting this issue. Use of yolov5 models are still being verified with EPD on our side.

Please ensure that EPD is running in Visualize mode.

If possible, please provide a way to reproduce this issue. Thank you.

@Sahil-cmd @mercedes1422 @carlowiesse Please have a look at this issue once you have the bandwidth.

cardboardcode commented 2 years ago

Hi @Prabhakaran-Gokul, may I clarify if you are using EPD docker which is activated when using the EPD Graphical User Interface (GUI) Deploy feature?

If so, there is a quick fix for this since it may narrow down the source of error to the use of docker.

cardboardcode commented 2 years ago

Closing due to staleness for more than 30 days. Feel free to reopen if there is still interest in resolving this issue.