Forums - Converted MobilenetSSD dlc produces only one output detection_classes:0.raw

3 posts / 0 new
or Register
Last post
Converted MobilenetSSD dlc produces only one output detection_classes:0.raw
hongbing.hu
Join Date: 31 May 18
Posts: 4
Posted: Wed, 2018-06-06 15:21
Top

Converted a sample MoibleNetSSD mode to a dlc file following the tutorial at https://developer.qualcomm.com/docs/snpe/convert_mobilenetssd.html. Three output nodes (detection_classes, detection_boxes, and detection_scores) are specified in the coversion command line as follows:

snpe-tensorflow-to-dlc --graph ssd_mobilenet_v1_coco_2017_11_17/frozen_inference_graph.pb -i Preprocessor/sub 300,300,3 --out_node detection_classes --out_node detection_boxes --out_node detection_scores --dlc mobilenet_ssd.dlc --allow_unconsumed_nodes

After executing spne-net-run with the converted dlc file, the output only contains one raw file (detection_classes:0.raw).  

snpe-net-run --container mobilenet_ssd.dlc --input_list data/mobilenet_file_list.txt

Should we expect two more output files for the other output nodes (detection_boxes, and detection_scroes) ?  

Thanks a lot.

 

 

  • Up0
  • Down0
Re: Converted MobilenetSSD dlc produces only one output...
jihoonk
Profile picture
Join Date: 28 Jan 13
Location: Seoul
Posts: 55
Posted: Wed, 2018-06-06 16:27
Top
Hi hongbing.hu,
snpe-net-run generates only the last layer's output by default. If you need other layers', there are two options. First one is to use "--debug" option, which will dump all layers' output. The other one is to specify layer names in input list file. For the details, refer to below reference page.

https://developer.qualcomm.com/docs/snpe/tools.html#tools_snpe-net-run

Thanks,

Jihoon

  • Up0
  • Down0
Re: Converted MobilenetSSD dlc produces only one output...
hongbing.hu
Join Date: 31 May 18
Posts: 4
Posted: Wed, 2018-06-06 18:10
Top

Hi Jihoonk,

Thanks a lot for your reply. Specifying the layer names in the input list works when using the mobileNet-SSD model in ssd_mobilenet_v1_coco_2017_11_17.tar.gz as described in the tutorial.

However, when using a newer Mobilenet v2 based model, the conversion tool snpe-tensorflow-to-dlc freezes in the middle of the conversion and will not produce any output. Can you advise what could be the reason? Is the latest Mobilenet v2 supported by the tool?

Thanks,

  • 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.