[MMSIG-80] Update and refine get_flops.py

[xin-li-67]

Motivation

MMSIG-80: update and refine get_flops.py

Modification

tools/analysis_tools/get_flops.py

BC-breaking (Optional)

Use cases (Optional)

Checklist

Before PR:

• I have read and followed the workflow indicated in the CONTRIBUTING.md to create this PR.
• Pre-commit or linting tools indicated in CONTRIBUTING.md are used to fix the potential lint issues.
• Bug fixes are covered by unit tests, the case that causes the bug should be added in the unit tests.
• New functionalities are covered by complete unit tests. If not, please add more unit tests to ensure correctness.
• The documentation has been modified accordingly, including docstring or example tutorials.

After PR:

• CLA has been signed and all committers have signed the CLA in this PR.

[codecov]

Codecov Report

Patch and project coverage have no change.

Comparison is base (896e9d5) 82.25% compared to head (4e16835) 82.26%.

❗ Current head 4e16835 differs from pull request most recent head f21d3a2. Consider uploading reports for the commit f21d3a2 to get more accurate results

Additional details and impacted files

@@            Coverage Diff            @@
##           dev-1.x    #2237    +/-   ##
=========================================
  Coverage    82.25%   82.26%            
=========================================
  Files          228      230     +2     
  Lines        13387    13552   +165     
  Branches      2268     2301    +33     
=========================================
+ Hits         11011    11148   +137     
- Misses        1862     1874    +12     
- Partials       514      530    +16     

Flag	Coverage Δ
unittests	82.26% <ø> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

see 21 files with indirect coverage changes

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[Ben-Louis]

Perhaps the argument can be converted to a boolean type using the store_true action? Additionally, the option '-c' may no longer be intuitive, as the argument name has been altered.

[xin-li-67]

Perhaps the argument can be converted to a boolean type using the store_true action? Additionally, the option '-c' may no longer be intuitive, as the argument name has been altered.

agree! I will update this in the next commit

[Tau-J]

1. --batch-input and --batch-size seems to be cumbersome. How about use batch_input = args.batch_size > 1?
2. I found a bug in mmengine: when inputs=None, get_model_complexity_info() cannot inference with cuda:0. I suggest setting cpu as default device for the moment. I'll report this bug to mmengine, or maybe someone will raise a PR to fix it.
3. The message is too long when printing both out_arch and out_table. I suggest only printing out_table and leaving out_arch as an optional argument.

[xin-li-67]

1. --batch-input and --batch-size seems to be cumbersome. How about use batch_input = args.batch_size > 1?
2. I found a bug in mmengine: when inputs=None, get_model_complexity_info() cannot inference with cuda:0. I suggest setting cpu as default device for the moment. I'll report this bug to mmengine, or maybe someone will raise a PR to fix it.
3. The message is too long when printing both out_arch and out_table. I suggest only printing out_table and leaving out_arch as an optional argument.

I made some changes in the latest commit:

1. removed batch-input and use batch-size;
2. now default device is CPU;
3. now out-table will be printed by default, and out-arch will not be printed unless the user specifies with --show-arch-info. However, it seems that the get_complexity_info function in MMEngine still needs to be refined, some models in MMPose are not well supported.

[Tau-J]

LGTM