Skip to content

[WIP] Dev #60

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.

Sign up for GitHub

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

Jump to bottom
Merged
merged 59 commits into from
Jun 4, 2020
Merged

[WIP] Dev #60

merged 59 commits into from
Jun 4, 2020

Conversation

bonlime
Copy link
Owner

@bonlime bonlime commented Apr 29, 2020

No description provided.

bonlime and others added 30 commits April 29, 2020 20:30
@bonlime
upd version
18b3db8
@bonlime
...
1ed4daf
@bonlime
add smooth l1 loss
c283581
@bonlime
support generate targets for batches
8efe866
@bonlime
move timer inside callback
2ae009d
@bonlime
add function to convert to weight standartization conv
e603637
@bonlime
add box_area function
01c3b1f
@bonlime
fix ws
a5c267f
@bonlime
add metrics reduction in FileLogger. Check mode in PhasesScheduler
4a48f97
@bonlime
add group norm and GN + WS model weights
ee199e2
@bonlime
minor update to README and docs
53e3a53
@bonlime
change assert to warning
ee7e1a7
@bonlime
fix bug for RMSprop
ee41bdc
@bonlime
support output stride in decoder. remove initialize for decoders
c906cf2
@bonlime
fix bug with biad in WS
5f37b99
@bonlime
add tests for WS
128f074
@bonlime
remove pool type selection (makes it ~1% faster)
7cf13d7
@bonlime
add polunomial decay to PhasesScheduler
247e725
@bonlime
add ModelEMA
5b267cc
@bonlime
make BlurPool use GaussBlur instead of AvgBlur by default
c4479d4
@bonlime
add Efficient Channel Attention
c716834
@bonlime
use as criterion in benchmarking
0df40e2
@bonlime
change deep_stem to stem_type (#64)
3be207b
@bonlime
Merge pull request #63 from bonlime/add_eca
6bf01cf 
add Efficient Channel Attention
@bonlime
update doc for ModelEma Callback
a991342
@bonlime
Merge branch 'dev' of https://github.com/bonlime/pytorch-tools into dev
99d02d2
@zakajd
fix encoder_features for efficientnet
7271b4a
@zakajd
add TF same Conv and MaxPool
72a5af7
@zakajd
add working EffDet
f46b341
@zakajd
minor
d3500e6
@bonlime
Copy link
Owner Author

bonlime commented May 21, 2020

Все еще нужно добавить тестов и примеров реального инференса для EffDet. Пока его невозможно использовать

bonlime and others added 24 commits May 21, 2020 20:39
@bonlime @zakajd
add bit-M-ResNet (#67)
be17775 
Co-authored-by: zakajd <[email protected]>
@zakajd
fix typo
b25b4f7
@zakajd
add accumulate batches
5ab7368
@zakajd
add tests. move loss scale to preserve loss for logging
305fc4d
@bonlime
Merge pull request #68 from bonlime/add_accumulate_batches
5f09e8a 
add accumulate batches
@zakajd
make swish & mish inplace
c38d97b
@zakajd
fully working EffDet + tests + pretrained weigths loading
0bf0a7c
@zakajd
format with black
51caf19
@zakajd
fix tests to use TF same padding
c2883a1
@zakajd
make swish & mish not inplace again. inplace version is incorrect
655d6ea
@bonlime @zakajd
Fully working RetinaNet (#69)
9951c85 
* add Activated No Norm (which is a simple activation)

* fully working Retina matching both mmdetection and TF Tpu versions

* add & update docs

* allow loading weights in RetinaNet

* add tests

* fix typos

Co-authored-by: zakajd <[email protected]>
@zakajd
fix initialization of EffNet models
aea23a1
@zakajd
fix same MaxPool
1258c2c
@zakajd
patch BN in effnet by default
22c6b1f
@zakajd
change init of last layer for Unet & HRNet
defbab1
@zakajd
add TF-like RMSprop
54a9131
@zakajd
also reduce loss_meter to save based on reduced loss
7c7e331
@zakajd
renaming and minor changes
be43c0e
@zakajd
rework some box utils and cover with tests
1e69ec3
@zakajd
make same ops scriptable
56e0047
@zakajd
fully matching EffDet
e2d35d8
@zakajd
simplify cat
19a2385
@bonlime
Merge pull request #70 from bonlime/fix_effdet
8222993 
Fully working EffDet
@zakajd
formatting
0568102
bonlime
bonlime commented Jun 4, 2020
View reviewed changes
Copy link
Owner Author

@bonlime bonlime left a comment

Choose a reason for hiding this comment

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

It's a dirty release. BiFPN segmentation is broken and EffDet is only half working. but otherwise, diff would be too large

@bonlime bonlime merged commit e73ec1f into master Jun 4, 2020
@bonlime bonlime deleted the dev branch June 4, 2020 17:21
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@bonlime @zakajd