[Feature] Add NSGANetV2 Algorithm

configs/_base_/nas_backbones/nsga_mobilenetv3_supernet.py

@@ -0,0 +1,106 @@
+# search space
+arch_setting = dict(
+    kernel_size=[  # [min_kernel_size, max_kernel_size, step]
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+        [3, 7, 2],
+    ],
+    num_blocks=[  # [min_num_blocks, max_num_blocks, step]
+        [1, 1, 1],
+        [1, 1, 1],
+        [0, 1, 1],
+        [0, 1, 1],
+        [1, 1, 1],
+        [1, 1, 1],
+        [0, 1, 1],
+        [0, 1, 1],
+        [1, 1, 1],
+        [1, 1, 1],
+        [0, 1, 1],
+        [0, 1, 1],
+        [1, 1, 1],
+        [1, 1, 1],
+        [0, 1, 1],
+        [0, 1, 1],
+        [1, 1, 1],
+        [1, 1, 1],
+        [0, 1, 1],
+        [0, 1, 1],
+    ],
+    expand_ratio=[  # [min_expand_ratio, max_expand_ratio, step]
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [3, 6, 1],
+        [6, 6, 1],  # last layer
+    ],
+    num_out_channels=[  # [min_channel, max_channel, step]
+        [16, 16, 1],  # first layer
+        [24, 24, 1],
+        [24, 24, 1],
+        [24, 24, 1],
+        [24, 24, 1],
+        [40, 40, 1],
+        [40, 40, 1],
+        [40, 40, 1],
+        [40, 40, 1],
+        [80, 80, 1],
+        [80, 80, 1],
+        [80, 80, 1],
+        [80, 80, 1],
+        [112, 112, 1],
+        [112, 112, 1],
+        [112, 112, 1],
+        [112, 112, 1],
+        [160, 160, 1],
+        [160, 160, 1],
+        [160, 160, 1],
+        [160, 160, 1],
+        [1280, 1280, 1],  # last layer
+    ])
+
+input_resizer_cfg = dict(
+    input_sizes=[[192, 192], [208, 208], [224, 224], [256, 256]])
+
+nas_backbone = dict(
+    type='AttentiveMobileNetV3',
+    arch_setting=arch_setting,
+    out_indices=(20, ),
+    stride_list=[2, 1, 1, 1] * 3 + [1] * 4 + [2] + [1] * 3,
+    with_se_list=[False] * 4 + [True] * 4 + [False] * 4 + [True] * 8,
+    act_cfg_list=['ReLU'] * 9 + ['HSwish'] * 13,
+    conv_cfg=dict(type='OFAConv2d'),
+    norm_cfg=dict(type='DynamicBatchNorm2d', momentum=0.1))

configs/_base_/settings/cifar10_bs96_nsga.py

@@ -0,0 +1,49 @@
+# dataset settings
+dataset_type = 'mmcls.CIFAR10'
+data_preprocessor = dict(
+    type='mmcls.ClsDataPreprocessor',
+    num_classes=10,
+    # RGB format normalization parameters
+    mean=[125.307, 122.961, 113.8575],
+    std=[51.5865, 50.847, 51.255],
+    # loaded images are already RGB format
+    to_rgb=False)
+
+train_pipeline = [
+    dict(type='mmcls.RandomCrop', crop_size=32, padding=4),
+    dict(type='mmcls.RandomFlip', prob=0.5, direction='horizontal'),
+    dict(type='mmcls.Cutout', shape=16, pad_val=0, prob=1.0),
+    dict(type='mmcls.PackClsInputs'),
+]
+
+test_pipeline = [
+    dict(type='mmcls.PackClsInputs'),
+]
+
+train_dataloader = dict(
+    batch_size=96,
+    num_workers=5,
+    dataset=dict(
+        type=dataset_type,
+        data_prefix='data/cifar10',
+        test_mode=False,
+        pipeline=train_pipeline),
+    sampler=dict(type='mmcls.DefaultSampler', shuffle=True),
+    persistent_workers=True,
+)
+
+val_dataloader = dict(
+    batch_size=96,
+    num_workers=5,
+    dataset=dict(
+        type=dataset_type,
+        data_prefix='data/cifar10/',
+        test_mode=True,
+        pipeline=test_pipeline),
+    sampler=dict(type='mmcls.DefaultSampler', shuffle=False),
+    persistent_workers=True,
+)
+val_evaluator = dict(type='mmcls.Accuracy', topk=(1, ))
+
+test_dataloader = val_dataloader
+test_evaluator = val_evaluator

configs/nas/mmcls/bignas/attentive_mobilenet_supernet_32xb64_in1k.py

@@ -44,7 +44,7 @@
             loss_kl=dict(
                 preds_S=dict(recorder='fc', from_student=True),
                 preds_T=dict(recorder='fc', from_student=False)))),
-    mutators=dict(type='mmrazor.NasMutator'))
+    mutator=dict(type='mmrazor.NasMutator'))
 
 model_wrapper_cfg = dict(
     type='mmrazor.BigNASDDP',

configs/nas/mmcls/nsganetv2/nsganetv2_mobilenet_search_8xb256_cifar10.py

@@ -0,0 +1,44 @@
+_base_ = ['./nsganetv2_mobilenet_supernet_1xb96_cifar10.py']
+
+model = dict(norm_training=True)
+
+train_dataloader = dict(batch_size=256)
+val_dataloader = dict(batch_size=256)
+test_dataloader = val_dataloader
+
+param_scheduler = [dict(type='ConstantLR', factor=1.0)]
+
+train_cfg = dict(
+    _delete_=True,
+    type='mmrazor.NSGA2SearchLoop',
+    dataloader=_base_.val_dataloader,
+    evaluator=_base_.val_evaluator,
+    max_epochs=30,
+    num_candidates=50,
+    top_k=10,
+    num_mutation=25,
+    num_crossover=25,
+    mutate_prob=0.3,
+    constraints_range=dict(flops=(0., 330.)),
+    score_key='accuracy/top1',
+    predictor_cfg=dict(
+        type='mmrazor.MetricPredictor',
+        encoding_type='normal',
+        train_samples=2,
+        handler_cfg=dict(type='mmrazor.GaussProcessHandler')),
+    finetune_cfg=dict(
+        model=_base_.model,
+        train_dataloader=_base_.train_dataloader,
+        train_cfg=dict(by_epoch=True, max_epochs=2),
+        optim_wrapper=dict(
+            type='OptimWrapper',
+            optimizer=dict(
+                type='SGD',
+                lr=0.1,
+                momentum=0.9,
+                weight_decay=1e-4,
+                nesterov=True)),
+        param_scheduler=param_scheduler,
+        default_hooks=_base_.default_hooks,
+    ),
+)

configs/nas/mmcls/nsganetv2/nsganetv2_mobilenet_search_8xb256_in1k.py

@@ -0,0 +1,40 @@
+_base_ = ['./nsganetv2_mobilenet_supernet_8xb128_in1k.py']
+
+model = dict(norm_training=True)
+
+param_scheduler = [dict(type='ConstantLR', factor=1.0)]
+
+train_cfg = dict(
+    _delete_=True,
+    type='mmrazor.NSGA2SearchLoop',
+    dataloader=_base_.val_dataloader,
+    evaluator=_base_.val_evaluator,
+    max_epochs=4,
+    num_candidates=4,
+    top_k=2,
+    num_mutation=2,
+    num_crossover=2,
+    mutate_prob=0.1,
+    constraints_range=dict(flops=(0., 360.)),
+    score_key='accuracy/top1',
+    predictor_cfg=dict(
+        type='mmrazor.MetricPredictor',
+        encoding_type='normal',
+        train_samples=2,
+        handler_cfg=dict(type='mmrazor.GaussProcessHandler')),
+    finetune_cfg=dict(
+        model=_base_.model,
+        train_dataloader=_base_.train_dataloader,
+        train_cfg=dict(by_epoch=True, max_epochs=1),
+        optim_wrapper=dict(
+            type='OptimWrapper',
+            optimizer=dict(
+                type='SGD',
+                lr=0.025,
+                momentum=0.9,
+                weight_decay=3e-4,
+                nesterov=True)),
+        param_scheduler=param_scheduler,
+        default_hooks=_base_.default_hooks,
+    ),
+)

configs/nas/mmcls/nsganetv2/nsganetv2_mobilenet_supernet_1xb96_cifar10.py

@@ -0,0 +1,39 @@
+_base_ = [
+    'mmcls::_base_/default_runtime.py',
+    'mmcls::_base_/schedules/imagenet_bs2048.py',
+    'mmrazor::_base_/settings/cifar10_bs96_nsga.py',
+    'mmrazor::_base_/nas_backbones/nsga_mobilenetv3_supernet.py',
+]
+
+supernet = dict(
+    _scope_='mmrazor',
+    type='SearchableImageClassifier',
+    backbone=_base_.nas_backbone,
+    neck=dict(type='SqueezeMeanPoolingWithDropout', drop_ratio=0.2),
+    head=dict(
+        type='DynamicLinearClsHead',
+        num_classes=1000,
+        in_channels=1280,
+        loss=dict(
+            type='mmcls.LabelSmoothLoss',
+            num_classes=1000,
+            label_smooth_val=0.1,
+            mode='original',
+            loss_weight=1.0),
+        topk=(1, 5)),
+    input_resizer_cfg=_base_.input_resizer_cfg,
+    connect_head=dict(connect_with_backbone='backbone.last_mutable_channels'),
+)
+
+model = dict(
+    _scope_='mmrazor',
+    type='NSGANetV2',
+    architecture=supernet,
+    data_preprocessor=_base_.data_preprocessor,
+    mutator=dict(type='mmrazor.NasMutator'))
+
+find_unused_parameters = True
+
+default_hooks = dict(
+    checkpoint=dict(
+        type='CheckpointHook', interval=1, max_keep_ckpts=1, save_best='auto'))

configs/nas/mmcls/nsganetv2/nsganetv2_mobilenet_supernet_8xb128_in1k.py

@@ -0,0 +1,38 @@
+_base_ = [
+    'mmcls::_base_/default_runtime.py',
+    'mmrazor::_base_/settings/imagenet_bs2048_bignas.py',
+    'mmrazor::_base_/nas_backbones/nsga_mobilenetv3_supernet.py',
+]
+
+supernet = dict(
+    _scope_='mmrazor',
+    type='SearchableImageClassifier',
+    backbone=_base_.nas_backbone,
+    neck=dict(type='SqueezeMeanPoolingWithDropout', drop_ratio=0.2),
+    head=dict(
+        type='DynamicLinearClsHead',
+        num_classes=1000,
+        in_channels=1280,
+        loss=dict(
+            type='mmcls.LabelSmoothLoss',
+            num_classes=1000,
+            label_smooth_val=0.1,
+            mode='original',
+            loss_weight=1.0),
+        topk=(1, 5)),
+    input_resizer_cfg=_base_.input_resizer_cfg,
+    connect_head=dict(connect_with_backbone='backbone.last_mutable_channels'),
+)
+
+model = dict(
+    _scope_='mmrazor',
+    type='NSGANetV2',
+    architecture=supernet,
+    data_preprocessor=_base_.data_preprocessor,
+    mutator=dict(type='mmrazor.NasMutator'))
+
+find_unused_parameters = True
+
+default_hooks = dict(
+    checkpoint=dict(
+        type='CheckpointHook', interval=1, max_keep_ckpts=1, save_best='auto'))

configs/nas/mmcls/onceforall/ofa_mobilenet_supernet_32xb64_in1k.py

@@ -43,7 +43,7 @@
             loss_kl=dict(
                 preds_S=dict(recorder='fc', from_student=True),
                 preds_T=dict(recorder='fc', from_student=False)))),
-    mutators=dict(type='mmrazor.NasMutator'))
+    mutator=dict(type='mmrazor.NasMutator'))
 
 model_wrapper_cfg = dict(
     type='mmrazor.BigNASDDP',

configs/nas/mmcls/spos/spos_shufflenet_search_nsga2_predictor_8xb128_in1k.py

@@ -0,0 +1,22 @@
+_base_ = ['./spos_shufflenet_supernet_8xb128_in1k.py']
+
+model = dict(norm_training=True)
+
+train_cfg = dict(
+    _delete_=True,
+    type='mmrazor.NSGA2SearchLoop',
+    dataloader=_base_.val_dataloader,
+    evaluator=_base_.val_evaluator,
+    max_epochs=20,
+    num_candidates=50,
+    top_k=10,
+    num_mutation=25,
+    num_crossover=25,
+    mutate_prob=0.1,
+    constraints_range=dict(flops=(0., 360.)),
+    predictor_cfg=dict(
+        type='mmrazor.MetricPredictor',
+        encoding_type='normal',
+        train_samples=2,
+        handler_cfg=dict(type='mmrazor.GaussProcessHandler')),
+)

mmrazor/engine/__init__.py

@@ -3,14 +3,14 @@
 from .optimizers import SeparateOptimWrapperConstructor
 from .runner import (AutoSlimGreedySearchLoop, DartsEpochBasedTrainLoop,
                      DartsIterBasedTrainLoop, EvolutionSearchLoop,
-                     GreedySamplerTrainLoop, SelfDistillValLoop,
-                     SingleTeacherDistillValLoop, SlimmableValLoop,
-                     SubnetValLoop)
+                     GreedySamplerTrainLoop, NSGA2SearchLoop,
+                     SelfDistillValLoop, SingleTeacherDistillValLoop,
+                     SlimmableValLoop, SubnetValLoop)
 
 __all__ = [
     'SeparateOptimWrapperConstructor', 'DumpSubnetHook',
     'SingleTeacherDistillValLoop', 'DartsEpochBasedTrainLoop',
     'DartsIterBasedTrainLoop', 'SlimmableValLoop', 'EvolutionSearchLoop',
-    'GreedySamplerTrainLoop', 'EstimateResourcesHook', 'SelfDistillValLoop',
-    'AutoSlimGreedySearchLoop', 'SubnetValLoop'
+    'GreedySamplerTrainLoop', 'SubnetValLoop', 'EstimateResourcesHook',
+    'SelfDistillValLoop', 'NSGA2SearchLoop', 'AutoSlimGreedySearchLoop'
 ]

mmrazor/engine/runner/__init__.py

@@ -4,6 +4,7 @@
 from .distill_val_loop import SelfDistillValLoop, SingleTeacherDistillValLoop
 from .evolution_search_loop import EvolutionSearchLoop
 from .iteprune_val_loop import ItePruneValLoop
+from .nsganetv2_search_loop import NSGA2SearchLoop
 from .slimmable_val_loop import SlimmableValLoop
 from .subnet_sampler_loop import GreedySamplerTrainLoop
 from .subnet_val_loop import SubnetValLoop
@@ -12,5 +13,5 @@
     'SingleTeacherDistillValLoop', 'DartsEpochBasedTrainLoop',
     'DartsIterBasedTrainLoop', 'SlimmableValLoop', 'EvolutionSearchLoop',
     'GreedySamplerTrainLoop', 'SubnetValLoop', 'SelfDistillValLoop',
-    'ItePruneValLoop', 'AutoSlimGreedySearchLoop'
+    'NSGA2SearchLoop', 'ItePruneValLoop', 'AutoSlimGreedySearchLoop'
 ]