Closed rentainhe closed 3 years ago
Hi @rentainhe thank you for this PR. Could you please provide a little bit of context? It would be nice if you can edit the description an explain what cbam does and link the paper. It looks similar to Channel + Spatial SE. So I am wondering what's the improvement, I see we concatenate the features from two poolings, would love to have your summary about it.
Tests are missing, you can add a test here https://github.com/FrancescoSaverioZuppichini/glasses/blob/master/test/test_att.py
Hi @rentainhe thank you for this PR. Could you please provide a little bit of context? It would be nice if you can edit the description an explain what cbam does and link the paper. It looks similar to Channel + Spatial SE. So I am wondering what's the improvement, I see we concatenate the features from two poolings, would love to have your summary about it.
Tests are missing, you can add a test here https://github.com/FrancescoSaverioZuppichini/glasses/blob/master/test/test_att.py
OKKKK, wait a minute~
Merging #279 (4ecefee) into master (bbd8e7c) will increase coverage by
0.05%
. The diff coverage isn/a
.
@@ Coverage Diff @@
## master #279 +/- ##
==========================================
+ Coverage 97.23% 97.28% +0.05%
==========================================
Files 82 86 +4
Lines 2998 3056 +58
==========================================
+ Hits 2915 2973 +58
Misses 83 83
Impacted Files | Coverage Δ | |
---|---|---|
glasses/models/AutoModel.py | 78.72% <ø> (ø) |
|
glasses/models/base/__init__.py | 100.00% <ø> (ø) |
|
glasses/models/base/protocols.py | 100.00% <ø> (ø) |
|
glasses/models/utils/scaler.py | 100.00% <ø> (ø) |
|
glasses/nn/__init__.py | 100.00% <ø> (ø) |
|
glasses/nn/att/CBAM.py | 100.00% <ø> (ø) |
|
glasses/nn/att/ECA.py | 100.00% <ø> (ø) |
|
glasses/nn/att/__init__.py | 100.00% <ø> (ø) |
|
glasses/nn/att/se.py | 100.00% <ø> (ø) |
|
glasses/nn/att/utils.py | 100.00% <ø> (ø) |
|
... and 21 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 77d049c...4ecefee. Read the comment docs.
Trying to implement
Cbam
Module in glassesOriginal Paper
address: https://arxiv.org/abs/1807.06521
Short Explanation
Cbam Module
improvesSE Module
in these ways:global_avg_pool
result, but also theglobal_max_pool
result.Conv2d
Layer on themean
andmax
features of the spatial space.TODO