Home Forums Model Development Kit (MDK) Intermediate Layer Output

This topic contains 1 reply, has 2 voices, and was last updated by Arpine Soghoyan Arpine Soghoyan 2 months, 1 week ago.

Viewing 2 posts - 1 through 2 (of 2 total)
  • Author
    Posts
  • #2352

    Daniel Wagner
    Participant

    Hi,

    in the most recent 2803 MDK documentation (2.1.0.3 TensorFlow version) under 6.7 it says that a model can also output the results of intermediate layers, but the documentation is not entirely clear to me.

    I understand that before converting the model, I need to set
    “learning”: true (and “last_layer_out”: false for all major layers)
    in the *_dat.json file of the model and then convert the model.

    However, going through the provided Python scripts it’s unclear to me how exactly to retrieve the additional results of those layers when running inference? Also, do I need to adapt the *_model.json file as well, e.g. change the output height or width or anything like that? Or how to set the chip_output_shape of the model?

    Thanks for your help

    #2521
    Arpine Soghoyan
    Arpine Soghoyan
    Moderator

    Hi Daniel,

    According to description in the user guide, the setting is provided in the .json file. It can be set for each major layer. The result (output of intermediate layer) will only be available after training when you run eval.py script. And by definition, intermediate layer means the layers before last sub-layer of last major layer.

    “learning”: true,

    If you’d like to see the output during chip inference, I’ll email a reference example.

    Regards,
    Arpine

Viewing 2 posts - 1 through 2 (of 2 total)

You must be logged in to reply to this topic.