My solution is based on UNet with a deep supervision branch for empty mask classification.
- Model: UNet.
- Backbone: ResNet34 backbone with frozen batch-normalization.
- Preprocessing: training on random crops with
(512, 512)size, inference on(768, 768)size. - Augmentations: ShiftScaleRotate, RandomBrightnessContrast, ElasticTransform, HorizontalFlip from albumentations.
- Optimizer: Adam, batch_size=8
- Scheduler: CosineAnnealingLR
- Additional feature: the proportion of non-empty samples linearly decreased from 0.8 to 0.22 (as in train dataset) depending on the epoch. It helped to converge faster.
- Loss:
2.7 * BCE(pred_mask, gt_mask) + 0.9 * DICE(pred_mask, gt_mask) + 0.1 * BCE(pred_empty, gt_empty). Herepred_maskis the prediction of the UNet, pred_empty is the prediction of the branch for empty mask classification. - Postprocessing:
if pred_empty > 0.4 or area(pred_mask) < 800: pred_mask = empty.Parameters are selected on the validation set. - Ensemble: averaging the 4 best checkpoints over 8 folds, horizontal flip TTA.
- Hardware: 4 x RTX 2080
make build
make run
make exec
cd scripts
bash train.sh
bash test.sh
bash ensemble.sh
bash submit.sh