Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

how to use in mmdet build model #63

Closed
791136190 opened this issue Mar 29, 2022 · 10 comments
Closed

how to use in mmdet build model #63

791136190 opened this issue Mar 29, 2022 · 10 comments
Labels

Comments

@791136190
Copy link

791136190 commented Mar 29, 2022

when use mmdet build this model, it will like:
object {
module list aaaa
module list bbb
}
when use prepare_by_platform to trace will get error like: TypeError: 'xxxobject' object does not support indexing

@791136190 791136190 changed the title how how to use in mmdet build model Mar 29, 2022
@Tracin
Copy link
Contributor

Tracin commented Mar 29, 2022

For object detection task, we highly recommand using EOD.
Which model do you want to do QAT on ?

@LMerCy
Copy link

LMerCy commented Mar 31, 2022

@Tracin Actually we also have requiremetns to integrate mqbench into mmdet, can eod be seen as a good practice?

@791136190
Copy link
Author

For model detection task, we highly recommand using EOD. Which model do you want to do QAT on ?

Since we are building our own model with mmdet, it may not be too convenient to use eod.

@Tracin
Copy link
Contributor

Tracin commented Mar 31, 2022

@Tracin Actually we also have requiremetns to integrate mqbench into mmdet, can eod be seen as a good practice?

Model in EOD is "trace friendly" it is important for MQBench to do qat.
If you have to use mmdet, a lot modification have to be done.
You have to make the model pass symbolic_trace smoothly.

@Tracin
Copy link
Contributor

Tracin commented Mar 31, 2022

For model detection task, we highly recommand using EOD. Which model do you want to do QAT on ?

Since we are building our own model with mmdet, it may not be too convenient to use eod.

If you have to use mmdet, a lot modification have to be done.
You have to make the model pass symbolic_trace smoothly.

@791136190
Copy link
Author

For model detection task, we highly recommand using EOD. Which model do you want to do QAT on ?

Since we are building our own model with mmdet, it may not be too convenient to use eod.

If you have to use mmdet, a lot modification have to be done. You have to make the model pass symbolic_trace smoothly.

This is a pity, because we are a chip factory, and we cannot change the training code.

@www516717402
Copy link
Contributor

www516717402 commented Apr 2, 2022

@791136190

  1. prepare_by_platform only for model. Cant use mmdet XXXDetector class
  2. I have done the translate by MQBench-0.04 and MMdet-2.21. you can ask when you meet any issue. (sorry for code cant share).

@791136190
Copy link
Author

@791136190

  1. prepare_by_platform only for model. Cant use mmdet XXXDetector class
  2. I have done the translate by MQBench-0.04 and MMdet-2.21. you can ask when you meet any issue. (sorry for code cant share).

@www516717402 Hello, can I take the liberty to ask how the basic modification process is? need modify the model build?

@www516717402
Copy link
Contributor

@791136190

  • Need rebuild Detector and head module.
  • Suggest put calib block before distribute model.

@github-actions
Copy link

github-actions bot commented Aug 1, 2022

This issue has not received any updates in 120 days. Please reply to this issue if this still unresolved!

@github-actions github-actions bot added the Stale label Aug 1, 2022
@github-actions github-actions bot closed this as completed Aug 9, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

No branches or pull requests

4 participants