Forums - Convert and Run for Faster RCNN and YOLOv3 model

2 posts / 0 new
or Register
Last post
Convert and Run for Faster RCNN and YOLOv3 model
aizat172
Join Date: 17 Aug 20
Posts: 1
Posted: Tue, 2021-08-03 23:22
Top

Hi guys,

I need help on how to convert and run the Faster RCNN and YOLO model on the SD chipset.

For FRCNN. Inside the documentation, it states that these type of model has been supported starting from v1.4.0 but there is no tutorial on how convert and run the model.

These are the links that I have been using for Faster RCNN setup

https://github.com/BVLC/caffe

Merged with https://github.com/rbgirshick/py-faster-rcnn

 

For YOLOv3 to DLC, what is the step to convert from YOLOv3 ONNX model to DLC model.

 

Thanks.

  • Up0
  • Down0
Re: Convert and Run for Faster RCNN and YOLOv3 model
SahilBandar
Join Date: 23 May 18
Posts: 37
Posted: Wed, 2021-08-11 09:16
Top

Hello aizat172,

You might have issues with respect to YoloV3 ONNX Models for some operations or unsupported type, below is the approach which we tried to make it work out.

  1. Download the Original Model of YoloV3 from Darknet's Official Website. (https://pjreddie.com/darknet/yolo/)
  2. Define the architecture of YoloV3 Model in Keras.
  3. Make sure that You are using Tensorflow 1.13 as Backend. There are some API level changes in latest Tensorflow which are not yet supported by SNPE.
  4. Reassign all weights of YoloV3.weights file to the YoloV3's defined architecture.
  5. Save the model back in Frozen Graph file format. (.pb file)
  6. Now use snpe-tensorflow-to-dlc tool for converting the model.

 

  • Up0
  • Down0
or Register

Opinions expressed in the content posted here are the personal opinions of the original authors, and do not necessarily reflect those of Qualcomm Incorporated or its subsidiaries (“Qualcomm”). The content is provided for informational purposes only and is not meant to be an endorsement or representation by Qualcomm or any other party. This site may also provide links or references to non-Qualcomm sites and resources. Qualcomm makes no representations, warranties, or other commitments whatsoever about any non-Qualcomm sites or third-party resources that may be referenced, accessible from, or linked to this site.