|
|
@@ -1,5 +1,5 @@
|
|
|
-batch_size: 2
|
|
|
-iters: 80000
|
|
|
+batch_size: 4
|
|
|
+iters: 160000
|
|
|
|
|
|
train_dataset:
|
|
|
type: Dataset
|
|
|
@@ -8,16 +8,16 @@ train_dataset:
|
|
|
num_classes: 19
|
|
|
transforms:
|
|
|
- type: ResizeStepScaling
|
|
|
- min_scale_factor: 0.5
|
|
|
- max_scale_factor: 2.0
|
|
|
- scale_step_size: 0.25
|
|
|
+ min_scale_factor: 0.125
|
|
|
+ max_scale_factor: 1.5
|
|
|
+ scale_step_size: 0.125
|
|
|
- type: RandomPaddingCrop
|
|
|
- crop_size: [512, 512]
|
|
|
+ crop_size: [1024, 512]
|
|
|
- type: RandomHorizontalFlip
|
|
|
- type: RandomDistort
|
|
|
- brightness_range: 0.4
|
|
|
- contrast_range: 0.4
|
|
|
- saturation_range: 0.4
|
|
|
+ brightness_range: 0.5
|
|
|
+ contrast_range: 0.5
|
|
|
+ saturation_range: 0.5
|
|
|
- type: Normalize
|
|
|
mode: train
|
|
|
|
|
|
@@ -39,17 +39,26 @@ model:
|
|
|
optimizer:
|
|
|
type: SGD
|
|
|
momentum: 0.9
|
|
|
- weight_decay: 4.0e-5
|
|
|
+ weight_decay: 5.0e-4
|
|
|
|
|
|
lr_scheduler:
|
|
|
type: PolynomialDecay
|
|
|
- learning_rate: 0.01
|
|
|
+ learning_rate: 0.005
|
|
|
end_lr: 0
|
|
|
power: 0.9
|
|
|
+ warmup_iters: 1000
|
|
|
+ warmup_start_lr: 1.0e-5
|
|
|
|
|
|
loss:
|
|
|
types:
|
|
|
- - type: CrossEntropyLoss
|
|
|
- - type: CrossEntropyLoss
|
|
|
- - type: CrossEntropyLoss
|
|
|
+ - type: OhemCrossEntropyLoss
|
|
|
+ min_kept: 130000 # batch_size * 1024 * 512 // 16
|
|
|
+ - type: OhemCrossEntropyLoss
|
|
|
+ min_kept: 130000
|
|
|
+ - type: OhemCrossEntropyLoss
|
|
|
+ min_kept: 130000
|
|
|
coef: [1, 1, 1]
|
|
|
+
|
|
|
+test_config:
|
|
|
+ aug_eval: True
|
|
|
+ scales: 0.5
|
Zhang Zelun