configs: update classification config files

configs/_base_/default_runtime_cls.py

@@ -29,7 +29,7 @@
 
 # set visualizer
 vis_backends = [dict(type='LocalVisBackend')]
-# visualizer = dict(type='mmcls.UniversalVisualizer', vis_backends=vis_backends)
+visualizer = dict(type='mmcls.ClsVisualizer', vis_backends=vis_backends)
 
 # set log level
 log_level = 'INFO'

configs/classification/mobnetv2_0.35_rep_1bx16_300e_cifar10.py

@@ -10,32 +10,34 @@
 data_root = 'datasets'
 height = 32
 width = 32
-batch_size = 16
-workers = 1
+batch_size = 32
+workers = 8
 persistent_workers = True
 
 # optimizer
 lr = 0.01
-epochs = 300
+epochs = 100
 
 
 model = dict(
  type='edgelab.ImageClassifier',
- backbone=dict(type='MobileNetv2', widen_factor=0.35, out_indices=(2,), rep=True),
+ data_preprocessor=dict(type='mmdet.DetDataPreprocessor', mean=[0.0, 0.0, 0.0], std=[255.0, 255.0, 255.0]),
+ backbone=dict(type='MobileNetv2', gray_input=False, widen_factor=0.35, out_indices=(2,), rep=True),
  neck=dict(type='mmcls.GlobalAveragePooling'),
  head=dict(
  type='mmcls.LinearClsHead',
- in_channels=16,
+ in_channels=32,
  num_classes=num_classes,
  loss=dict(type='mmcls.CrossEntropyLoss', loss_weight=1.0),
  topk=(1, 5),
  ),
 )
 
 train_pipeline = [
+ dict(type='mmengine.Resize', scale=(height, width)),
+ dict(type='mmcls.ColorJitter', brightness=0.2, contrast=0.2, saturation=0.2, hue=0.2),
  dict(type='mmcls.Rotate', angle=30.0, prob=0.6),
  dict(type='mmcls.RandomFlip', prob=0.5, direction='horizontal'),
- dict(type='mmengine.Resize', scale=(height, width)),
  dict(type='mmcls.PackClsInputs'),
 ]
 
@@ -93,12 +95,12 @@
  type='MultiStepLR',
  begin=1,
  end=500,
- milestones=[100, 200, 250],
+ milestones=[30, 60, 90],
  gamma=0.1,
  by_epoch=True,
  ),
 ]
 
 auto_scale_lr = dict(base_batch_size=batch_size)
 
-train_cfg = dict(by_epoch=True, max_epochs=epochs)
+train_cfg = dict(by_epoch=True, max_epochs=epochs, val_interval=5)

configs/classification/mobnetv2_0.35_rep_1bx16_300e_cifar100.py

@@ -10,17 +10,18 @@
 data_root = 'datasets'
 height = 32
 width = 32
-batch_size = 16
-workers = 1
+batch_size = 32
+workers = 8
 persistent_workers = True
 
 # optimizer
 lr = 0.01
-epochs = 300
+epochs = 100
 
 
 model = dict(
  type='edgelab.ImageClassifier',
+ data_preprocessor=dict(type='mmdet.DetDataPreprocessor', mean=[0.0, 0.0, 0.0], std=[255.0, 255.0, 255.0]),
  backbone=dict(type='MobileNetv2', widen_factor=0.35, out_indices=(2,), rep=True),
  neck=dict(type='mmcls.GlobalAveragePooling'),
  head=dict(
@@ -33,9 +34,10 @@
 )
 
 train_pipeline = [
+ dict(type='mmengine.Resize', scale=(height, width)),
+ dict(type='mmcls.ColorJitter', brightness=0.2, contrast=0.2, saturation=0.2, hue=0.2),
  dict(type='mmcls.Rotate', angle=30.0, prob=0.6),
  dict(type='mmcls.RandomFlip', prob=0.5, direction='horizontal'),
- dict(type='mmengine.Resize', scale=(height, width)),
  dict(type='mmcls.PackClsInputs'),
 ]
 
@@ -93,12 +95,12 @@
  type='MultiStepLR',
  begin=1,
  end=500,
- milestones=[100, 200, 250],
+ milestones=[30, 60, 90],
  gamma=0.1,
  by_epoch=True,
  ),
 ]
 
 auto_scale_lr = dict(base_batch_size=batch_size)
 
-train_cfg = dict(by_epoch=True, max_epochs=epochs)
+train_cfg = dict(by_epoch=True, max_epochs=epochs, val_interval=5)

configs/classification/mobnetv2_0.35_rep_1bx16_300e_custom.py

@@ -10,24 +10,18 @@
 data_root = 'datasets/digit'
 height = 96
 width = 96
-batch_size = 16
-workers = 1
+batch_size = 32
+workers = 8
 persistent_workers = True
 
 # optimizer
 lr = 0.01
-epochs = 300
+epochs = 100
 
-data_preprocessor = dict(
- type='mmcls.ClsDataPreprocessor',
- mean=[0, 0, 0],
- std=[255.0, 255.0, 255.0],
- to_rgb=True,
-)
 
 model = dict(
  type='edgelab.ImageClassifier',
- data_preprocessor=data_preprocessor,
+ data_preprocessor=dict(type='mmdet.DetDataPreprocessor', mean=[0.0, 0.0, 0.0], std=[255.0, 255.0, 255.0]),
  backbone=dict(type='MobileNetv2', widen_factor=0.35, out_indices=(2,), rep=True),
  neck=dict(type='mmcls.GlobalAveragePooling', dim=2),
  head=dict(
@@ -42,9 +36,10 @@
 
 train_pipeline = [
  dict(type='LoadImageFromFile'),
+ dict(type='mmengine.Resize', scale=(height, width)),
+ dict(type='mmcls.ColorJitter', brightness=0.2, contrast=0.2, saturation=0.2, hue=0.2),
  dict(type='mmcls.Rotate', angle=30.0, prob=0.6),
  dict(type='mmcls.RandomFlip', prob=0.5, direction='horizontal'),
- dict(type='mmengine.Resize', scale=(height, width)),
  dict(type='mmcls.PackClsInputs'),
 ]
 
@@ -100,12 +95,12 @@
  type='MultiStepLR',
  begin=1,
  end=500,
- milestones=[100, 200, 250],
+ milestones=[30, 60, 90],
  gamma=0.1,
  by_epoch=True,
  ),
 ]
 
 auto_scale_lr = dict(base_batch_size=batch_size)
 
-train_cfg = dict(by_epoch=True, max_epochs=epochs)
+train_cfg = dict(by_epoch=True, max_epochs=epochs, val_interval=5)

configs/classification/mobnetv2_0.35_rep_1bx16_300e_mnist.py

@@ -0,0 +1,106 @@
+_base_ = '../_base_/default_runtime_cls.py'
+default_scope = 'edgelab'
+custom_imports = dict(imports=['edgelab'], allow_failed_imports=False)
+
+# model settings
+num_classes = 10
+gray = True
+
+# dataset settings
+dataset_type = 'mmcls.MNIST'
+data_root = 'datasets'
+height = 32
+width = 32
+batch_size = 32
+workers = 8
+persistent_workers = True
+
+# optimizer
+lr = 0.01
+epochs = 100
+
+
+model = dict(
+ type='edgelab.ImageClassifier',
+ data_preprocessor=dict(type='mmdet.DetDataPreprocessor', mean=[0.0], std=[255.0]),
+ backbone=dict(type='MobileNetv2', gray_input=gray, widen_factor=0.35, out_indices=(2,), rep=True),
+ neck=dict(type='mmcls.GlobalAveragePooling'),
+ head=dict(
+ type='mmcls.LinearClsHead',
+ in_channels=32,
+ num_classes=num_classes,
+ loss=dict(type='mmcls.CrossEntropyLoss', loss_weight=1.0),
+ topk=(1, 5),
+ ),
+)
+
+train_pipeline = [
+ dict(type='mmengine.Resize', scale=(height, width)),
+ dict(type='mmcls.Rotate', angle=30.0, prob=0.6),
+ dict(type='mmcls.RandomFlip', prob=0.5, direction='horizontal'),
+ dict(type='mmcls.PackClsInputs'),
+]
+
+test_pipeline = [
+ dict(type='mmengine.Resize', scale=(height, width)),
+ dict(type='mmcls.PackClsInputs'),
+]
+
+train_dataloader = dict(
+ # Training dataset configurations
+ batch_size=batch_size,
+ num_workers=workers,
+ persistent_workers=persistent_workers,
+ dataset=dict(
+ type=dataset_type,
+ data_root=data_root,
+ data_prefix='cifar10/',
+ test_mode=False,
+ pipeline=train_pipeline,
+ ),
+ sampler=dict(type='DefaultSampler', shuffle=True),
+)
+
+val_dataloader = dict(
+ batch_size=batch_size,
+ num_workers=workers,
+ persistent_workers=persistent_workers,
+ dataset=dict(
+ type=dataset_type,
+ data_root=data_root,
+ data_prefix='cifar10/',
+ test_mode=True,
+ pipeline=test_pipeline,
+ ),
+ sampler=dict(type='DefaultSampler', shuffle=False),
+)
+
+test_dataloader = val_dataloader
+
+# evaluator
+val_evaluator = dict(type='mmcls.Accuracy', topk=(1, 5))
+test_evaluator = val_evaluator
+
+
+val_cfg = dict()
+test_cfg = dict()
+
+# optimizer
+optim_wrapper = dict(optimizer=dict(type='SGD', lr=lr, momentum=0.9, weight_decay=0.0001))
+
+# learning policy
+param_scheduler = [
+ dict(type='LinearLR', begin=0, end=30, start_factor=0.001, by_epoch=False), # warm-up
+ dict(
+ type='MultiStepLR',
+ begin=1,
+ end=500,
+ milestones=[30, 60, 90],
+ gamma=0.1,
+ by_epoch=True,
+ ),
+]
+
+auto_scale_lr = dict(base_batch_size=batch_size)
+
+train_cfg = dict(by_epoch=True, max_epochs=epochs, val_interval=5)

configs/classification/mobnetv2_1.0_1bx16_300e_cifar10.py

@@ -10,16 +10,17 @@
 data_root = 'datasets'
 height = 32
 width = 32
-batch_size = 16
-workers = 1
+batch_size = 32
+workers = 8
 persistent_workers = True
 
 # optimizer
 lr = 0.01
-epochs = 300
+epochs = 100
 
 model = dict(
  type='edgelab.ImageClassifier',
+ data_preprocessor=dict(type='mmdet.DetDataPreprocessor', mean=[0.0, 0.0, 0.0], std=[255.0, 255.0, 255.0]),
  backbone=dict(type='mmcls.MobileNetV2', widen_factor=1.0),
  neck=dict(type='mmcls.GlobalAveragePooling'),
  head=dict(
@@ -32,9 +33,10 @@
 )
 
 train_pipeline = [
+ dict(type='mmengine.Resize', scale=(height, width)),
+ dict(type='mmcls.ColorJitter', brightness=0.2, contrast=0.2, saturation=0.2, hue=0.2),
  dict(type='mmcls.Rotate', angle=30.0, prob=0.6),
  dict(type='mmcls.RandomFlip', prob=0.5, direction='horizontal'),
- dict(type='mmengine.Resize', scale=(height, width)),
  dict(type='mmcls.PackClsInputs'),
 ]
 
@@ -91,12 +93,12 @@
  type='MultiStepLR',
  begin=1,
  end=500,
- milestones=[100, 200, 250],
+ milestones=[30, 60, 90],
  gamma=0.1,
  by_epoch=True,
  ),
 ]
 
 auto_scale_lr = dict(base_batch_size=batch_size)
 
-train_cfg = dict(by_epoch=True, max_epochs=epochs)
+train_cfg = dict(by_epoch=True, max_epochs=epochs, val_interval=5)

configs/classification/mobnetv2_1.0_1bx16_300e_cifar100.py

@@ -10,16 +10,17 @@
 data_root = 'datasets'
 height = 32
 width = 32
-batch_size = 16
-workers = 1
+batch_size = 32
+workers = 8
 persistent_workers = True
 
 # optimizer
 lr = 0.01
-epochs = 300
+epochs = 100
 
 model = dict(
  type='edgelab.ImageClassifier',
+ data_preprocessor=dict(type='mmdet.DetDataPreprocessor', mean=[0.0, 0.0, 0.0], std=[255.0, 255.0, 255.0]),
  backbone=dict(type='mmcls.MobileNetV2', widen_factor=1.0),
  neck=dict(type='mmcls.GlobalAveragePooling'),
  head=dict(
@@ -32,9 +33,10 @@
 )
 
 train_pipeline = [
+ dict(type='mmengine.Resize', scale=(height, width)),
+ dict(type='mmcls.ColorJitter', brightness=0.2, contrast=0.2, saturation=0.2, hue=0.2),
  dict(type='mmcls.Rotate', angle=30.0, prob=0.6),
  dict(type='mmcls.RandomFlip', prob=0.5, direction='horizontal'),
- dict(type='mmengine.Resize', scale=(height, width)),
  dict(type='mmcls.PackClsInputs'),
 ]
 
@@ -91,12 +93,12 @@
  type='MultiStepLR',
  begin=1,
  end=500,
- milestones=[100, 200, 250],
+ milestones=[30, 60, 90],
  gamma=0.1,
  by_epoch=True,
  ),
 ]
 
 auto_scale_lr = dict(base_batch_size=batch_size)
 
-train_cfg = dict(by_epoch=True, max_epochs=epochs)
+train_cfg = dict(by_epoch=True, max_epochs=epochs, val_interval=5)