Skip to content

[TFLite] Using real image for QNN testing. - #4816

Merged
FrozenGene merged 5 commits into
apache:masterfrom
anijain2305:tflite_image
Feb 11, 2020
Merged

[TFLite] Using real image for QNN testing.#4816
FrozenGene merged 5 commits into
apache:masterfrom
anijain2305:tflite_image

Conversation

@anijain2305

@anijain2305anijain2305 commented Feb 4, 2020

Copy link
Copy Markdown
Contributor

This PR does 2 things

  • Test QNN on a real image. QNN and TFLite outputs differ due to rounding differences, making random input not a good candidate for label testing. So, this PR adds a real image for testing QNN.
  • @inadob also found a bug in QNN parsing [Frontend][TFLite] Fix quantized pad value for convolution #4807 This PR needs that fix as well to enable the real image tests, so adding that fix here.

Thanks to @inadob for finding the bug

@FrozenGene@kevinthesun@inadob Can you please review?

FrozenGene
FrozenGene reviewed Feb 5, 2020
View reviewed changes
Comment threadpython/tvm/relay/frontend/tflite.py
FrozenGene
FrozenGene requested changes Feb 5, 2020
View reviewed changes
Comment threadpython/tvm/relay/frontend/tflite.py
@anijain2305

anijain2305 commented Feb 5, 2020•
edited
Loading

Copy link
Copy Markdown
ContributorAuthor

@inadob Can you please review the TFLite pad test portion? I am not very familiar with using fake quant, and I just replicated your code for Pad. So, closer scrutiny would be good.

@anijain2305anijain2305 mentioned this pull request Feb 5, 2020
Closed
@u99127u99127 mentioned this pull request Feb 5, 2020
Closed
@anijain2305

anijain2305 commented Feb 6, 2020

Copy link
Copy Markdown
ContributorAuthor

Ping @FrozenGene

@FrozenGene

FrozenGene commented Feb 7, 2020

Copy link
Copy Markdown
Member

ping @inadob , would you review it again? If you have no other comments, would you set explicit approve? Thanks?

inadob
inadob suggested changes Feb 7, 2020
View reviewed changes

@inadobinadob left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Almost ready, have a couple of small changes below

Comment threadtests/python/frontend/tflite/test_forward.py
@anijain2305

anijain2305 commented Feb 9, 2020

Copy link
Copy Markdown
ContributorAuthor

@inadob Done

@anijain2305

anijain2305 commented Feb 10, 2020

Copy link
Copy Markdown
ContributorAuthor

@FrozenGene Good to merge?

@FrozenGene

FrozenGene commented Feb 11, 2020

Copy link
Copy Markdown
Member

Thanks @anijain2305@inadob It is merged now.

@FrozenGene
FrozenGene merged commit 902e21b into apache:masterFeb 11, 2020
@masahi

masahi commented Feb 12, 2020•
edited
Loading

Copy link
Copy Markdown
Member

What happen if the zero point is a vector, as in per channel quantization? What should the pad value be?

@u99127

u99127 commented Feb 20, 2020

Copy link
Copy Markdown

What happen if the zero point is a vector, as in per channel quantization? What should the pad value be?

@masahi good question but please file a separate issue only affecting master since per channel quantization came in after this. Otherwise I fear this will be missed.

Ramana

@u99127

u99127 commented Feb 20, 2020

Copy link
Copy Markdown

@anijain2305 - this needs a backport to 0.6 tag ?

@masahi

masahi commented Feb 20, 2020

Copy link
Copy Markdown
Member

@u99127 I figured the answer myself. QNN support for per channel quantization applies to only weight quantization (which is supposed to be a common restriction in other frameworks as well). For pad value we can use a scalar zero point associated with activations.

@anijain2305

anijain2305 commented Feb 20, 2020

Copy link
Copy Markdown
ContributorAuthor

@masahi You are right (as always :) )

@u99127 Let me look into this.

alexwong pushed a commit to alexwong/tvm that referenced this pull request Feb 26, 2020
@anijain2305@alexwong
[TFLite] Using real image for QNN testing. (apache#4816)
e518df2
* [TFLite] Using real image for QNN testing.
* Setting seed for SSD mobilenet for fixed input.
* Support quantized Pad op.
* Remove unnnecessary line.
* Ina comments.
alexwong pushed a commit to alexwong/tvm that referenced this pull request Feb 28, 2020
@anijain2305@alexwong
[TFLite] Using real image for QNN testing. (apache#4816)
7306114
* [TFLite] Using real image for QNN testing.
* Setting seed for SSD mobilenet for fixed input.
* Support quantized Pad op.
* Remove unnnecessary line.
* Ina comments.
zhiics pushed a commit to neo-ai/tvm that referenced this pull request Mar 2, 2020
@anijain2305@zhiics
[TFLite] Using real image for QNN testing. (apache#4816)
880b60f
* [TFLite] Using real image for QNN testing.
* Setting seed for SSD mobilenet for fixed input.
* Support quantized Pad op.
* Remove unnnecessary line.
* Ina comments.
@anijain2305anijain2305 mentioned this pull request Jun 26, 2020
Closed
@ZihengJiangZihengJiang mentioned this pull request Sep 17, 2020
Closed
Sign up for freeto join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@FrozenGeneFrozenGeneFrozenGene approved these changes
+1 more reviewer
@inadobinadobinadob approved these changes
Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

5 participants

@anijain2305@FrozenGene@masahi@u99127@inadob