Forums - Help!!!! Error in converting a pretrained ssd mobilenet V1 checkpoint to dlc

2 posts / 0 new
Last post
Help!!!! Error in converting a pretrained ssd mobilenet V1 checkpoint to dlc
tmeharizghi
Join Date: 12 Mar 18
Posts: 4
Posted: Tue, 2018-08-28 14:04

Hello everyone,

I downloaded a TensorFlow pretrained ssd_mobilenet_v1 object detection model based on Coco dataset from here:

http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v1_coco_2018_01_28.tar.gz

I could be able to convert the given frozen pb file to DLC format. But if I take the downloaded checkpoint, export and freeze it and finally convert to dlc, I could NOT be able to do that. I could get only a 1KB of DLC file.

Could someone try the following steps for me to convert to dlc and provide me a feedback it works on his/her side?

1. Download the pretrained model from the above link. And uncompress it.

2. Once it is uncompressed, export and freeze the given model checkpoint file to pb format by using 'export_inference_graph.py' from TensorFlow Object Detection API.

3. Finally convert the frozen graph to DLC using SNPE 'snpe-tensorflow-to-dlc' tool.

Please let me know if the above steps work and you get a valid DLC file.

Thank you in advance.

-Tes

  • Up0
  • Down0
gesqdn-forum
Join Date: 4 Nov 18
Posts: 184
Posted: Tue, 2019-09-17 05:09

Hi,

I have followed the steps you mentioned above and successfully able to get a .dlc file from downloaded checkpoints. 

The size of .pb file which is generated from checkpoints using 'export_inference_graph.py' of TensorFlow Object Detection API is 29.1MB and the .dlc file generated from the 'snpe-tensorflow-to-dlc' tool is 27.4MB.

SNPE version used : snpe-1.29.0.456

Thank you.

  • 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.