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

Feature Pyramid Grids #4645

Merged
merged 9 commits into from
Feb 26, 2021
Merged

Feature Pyramid Grids #4645

merged 9 commits into from
Feb 26, 2021

Conversation

yhcao6
Copy link
Collaborator

@yhcao6 yhcao6 commented Feb 20, 2021

Implementation of FPG that published in https://arxiv.org/abs/2004.03580.

Here is the mAP comparison of this PR and paper:

Model Paper This PR
Faster_R50_FPG_640 41.4 42.2
Faster_R50_FPG-cha128_640 40.0 41.2
Mask_R50_FPG_640 42.4 42.7
Mask_R50_FPG-cha128_640 40.9 41.7
Retinanet_R50_FPG_640 40.0 40.5
Retinanet_R50_FPG-cha128_640 39.0

@yhcao6 yhcao6 added the WIP Working in progress label Feb 20, 2021
@yhcao6 yhcao6 requested a review from hellock February 20, 2021 06:24
@codecov
Copy link

codecov bot commented Feb 20, 2021

Codecov Report

Merging #4645 (1059b48) into master (bf128b0) will decrease coverage by 0.13%.
The diff coverage is 51.49%.

Impacted file tree graph

@@            Coverage Diff             @@
##           master    #4645      +/-   ##
==========================================
- Coverage   64.17%   64.03%   -0.14%     
==========================================
  Files         250      252       +2     
  Lines       19286    19631     +345     
  Branches     3264     3323      +59     
==========================================
+ Hits        12377    12571     +194     
- Misses       6258     6393     +135     
- Partials      651      667      +16     
Flag Coverage Δ
unittests 64.00% <51.49%> (-0.14%) ⬇️

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

Impacted Files Coverage Δ
mmdet/core/mask/mask_target.py 95.83% <ø> (ø)
mmdet/datasets/builder.py 65.27% <0.00%> (ø)
mmdet/datasets/samplers/group_sampler.py 18.75% <0.00%> (-0.24%) ⬇️
mmdet/models/losses/cross_entropy_loss.py 100.00% <ø> (ø)
mmdet/models/roi_heads/mask_heads/fcn_mask_head.py 71.77% <ø> (ø)
mmdet/core/mask/structures.py 76.74% <6.77%> (-11.13%) ⬇️
mmdet/models/dense_heads/ga_rpn_head.py 27.27% <15.00%> (-1.50%) ⬇️
mmdet/models/dense_heads/corner_head.py 75.60% <20.00%> (-0.76%) ⬇️
mmdet/datasets/samplers/distributed_sampler.py 33.33% <25.00%> (ø)
mmdet/core/post_processing/merge_augs.py 70.00% <37.50%> (-10.36%) ⬇️
... and 27 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update bf128b0...59e8df8. Read the comment docs.

mmdet/models/necks/fpg.py Outdated Show resolved Hide resolved
mmdet/models/necks/fpg.py Outdated Show resolved Hide resolved
mmdet/models/necks/fpg.py Show resolved Hide resolved
mmdet/models/necks/fpg.py Outdated Show resolved Hide resolved
mmdet/models/necks/fpg.py Outdated Show resolved Hide resolved
norm_cfg (dict): Config dict for normalization layer. Default: None.
"""

transition_types = {
Copy link
Collaborator

Choose a reason for hiding this comment

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

Can we create a new registry for transition thus we could extend the transitions without modifying the code in fpg.py?

mmdet/models/necks/fpg.py Outdated Show resolved Hide resolved
mmdet/models/necks/fpg.py Show resolved Hide resolved
mmdet/models/necks/fpg.py Outdated Show resolved Hide resolved
self.out_channels = out_channels
self.num_ins = len(in_channels)
self.num_outs = num_outs
if inter_channels is None:
Copy link
Collaborator

Choose a reason for hiding this comment

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

Do we need a unit test to more comprehensively test this module?

configs/fpg/README.md Outdated Show resolved Hide resolved
mmdet/models/necks/fpg.py Outdated Show resolved Hide resolved
@ZwwWayne ZwwWayne merged commit e32ed8d into open-mmlab:master Feb 26, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
WIP Working in progress
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants