- pretrained models
- EfficientNet B7 (seed=42, https://github.com/lukemelas/EfficientNet-PyTorch)
- EfficientNet B7 Noisy Student (seed=43, https://github.com/rwightman/pytorch-image-models)
- EfficientNet B8 (seed=44, https://github.com/rwightman/pytorch-image-models)
- image size
- (256,256)
- DA
- HorizontalFlip
- RandAugument (N=2, M=15) (ttps://github.com/rwightman/pytorch-image-models)
- CV
- 5fold(StratifiedKFold)
- epoch
- 30
- optimizer
- SGD
- lr
- 最後の全結合層: 1e-1
- それ以外 : 1e-2
- weight_decay: 1e-4
- momentum: 0.9
- lr
- SGD
- scheduler
- 21エポック以降は学習率を0.5倍
- criterion
- CrossEntropyLoss
- TTA
- w/o HorizontalFlip
- w/ HorizontalFlip
- ensemble
- モデル3種とTTA2パターンの合計6つのモデルでアンサンブル
-
Notifications
You must be signed in to change notification settings - Fork 2
tealgreen0503/nishika-japanese-painting-competition
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
About
nishika japanese painting competition 2nd place solution
Resources
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published