Add Class-Incremental Learning features for Classification & Segmentation

constraints.txt

@@ -0,0 +1,11 @@
+mmcv-full>=1.3.1
+mmcls>=0.9.0
+# numpy>=1.21.1
+numpy==1.19.5
+pillow>=9.0.1
+networkx>=2.7
+scikit-image>=0.17.2
+# instaboostfast>=0.1.2
+# torch>=1.7.1,<=1.9.0
+# torchvision>=0.8.1,<=0.10.0
+pyparsing==2.4.7

models/classification/_base_/densenet_121_omz.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    pretrained=None,
+    backbone=dict(
+        type='OmzBackboneCls',
+        mode='train',
+        model_path='public/densenet-121-caffe2/FP32/densenet-121-caffe2.xml',
+        last_layer_name='conv5_blk/bn_3',
+        normalized_img_input=True
+    ),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1024,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/efficientnet_b0_omz.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    pretrained=None,
+    backbone=dict(
+        type='OmzBackboneCls',
+        mode='train',
+        model_path='public/efficientnet-b0/FP32/efficientnet-b0.xml',
+        last_layer_name='efficientnet-b0/model/blocks_15/tpu_batch_normalization_2/batchnorm/add_1',
+        normalized_img_input=True
+    ),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=320,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/efficientnet_b5_omz.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    pretrained=None,
+    backbone=dict(
+        type='OmzBackboneCls',
+        mode='train',
+        model_path='public/efficientnet-b5/FP32/efficientnet-b5.xml',
+        last_layer_name='efficientnet-b5/model/blocks_38/Add',
+        normalized_img_input=True
+    ),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=512,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/efficientnet_b7_omz.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    pretrained=None,
+    backbone=dict(
+        type='OmzBackboneCls',
+        mode='train',
+        model_path='public/efficientnet-b7_auto_aug/FP32/efficientnet-b7_auto_aug.xml',
+        last_layer_name='efficientnet-b7/model/blocks_54/Add',
+        normalized_img_input=True
+    ),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=640,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/googlenet_v1_omz.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    pretrained=None,
+    backbone=dict(
+        type='OmzBackboneCls',
+        mode='train',
+        model_path='public/googlenet-v1-tf/FP32/googlenet-v1-tf.xml',
+        last_layer_name='InceptionV1/InceptionV1/Mixed_5c/concat_v2',
+        normalized_img_input=True
+    ),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1024,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/googlenet_v2_omz.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    pretrained=None,
+    backbone=dict(
+        type='OmzBackboneCls',
+        mode='train',
+        model_path='public/googlenet-v2/FP32/googlenet-v2.xml',
+        last_layer_name='inception_5b/output',
+        normalized_img_input=True
+    ),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1024,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/googlenet_v3_omz.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    pretrained=None,
+    backbone=dict(
+        type='OmzBackboneCls',
+        mode='train',
+        model_path='public/googlenet-v3/FP32/googlenet-v3.xml',
+        last_layer_name='InceptionV3/InceptionV3/Mixed_7c/concat_v2',
+        normalized_img_input=True
+    ),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=2048,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/googlenet_v4_omz.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    pretrained=None,
+    backbone=dict(
+        type='OmzBackboneCls',
+        mode='train',
+        model_path='public/googlenet-v4-tf/FP32/googlenet-v4-tf.xml',
+        last_layer_name='InceptionV4/InceptionV4/Mixed_7d/concat',
+        normalized_img_input=True
+    ),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1536,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/mobilenet_v1_omz.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    pretrained=None,
+    backbone=dict(
+        type='OmzBackboneCls',
+        mode='train',
+        model_path='public/mobilenet-v1-1.0-224/FP32/mobilenet-v1-1.0-224.xml',
+        last_layer_name='relu6/sep',
+        normalized_img_input=True
+    ),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1024,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/mobilenet_v2.py

@@ -0,0 +1,11 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(type='MobileNetV2', widen_factor=1.0),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1280,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/mobilenet_v2_omz.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    pretrained=None,
+    backbone=dict(
+        type='OmzBackboneCls',
+        mode='train',
+        model_path='public/mobilenet-v2/FP32/mobilenet-v2.xml',
+        last_layer_name='relu6_4',
+        normalized_img_input=True
+    ),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1280,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/ote_efficientnet.py

@@ -0,0 +1,13 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(
+        type='OTEEfficientNet',
+        version='b0'),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1280,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/ote_mobilenet_v3.py

@@ -0,0 +1,16 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(
+        type='OTEMobileNetV3',
+        mode='small',
+        width_mult=1.0),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='NonLinearClsHead',
+        num_classes=1000,
+        in_channels=576,
+        hid_channels=1024,
+        act_cfg=dict(type='HSwish'),
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/resnet18.py

@@ -0,0 +1,17 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(
+        type='ResNet',
+        depth=18,
+        num_stages=4,
+        out_indices=(3, ),
+        style='pytorch'),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        in_channels=512,
+        num_classes=10,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        topk=(1, 5),
+    ))

models/classification/_base_/resnet34.py

@@ -0,0 +1,17 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(
+        type='ResNet',
+        depth=34,
+        num_stages=4,
+        out_indices=(3, ),
+        style='pytorch'),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        in_channels=512,
+        num_classes=10,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        topk=(1, 5),
+    ))

models/classification/_base_/resnet50.py

@@ -0,0 +1,18 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(
+        type='ResNet',
+        depth=50,
+        in_channels=3,
+        num_stages=4,
+        out_indices=(3, ),
+        norm_cfg=dict(type='BN')),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=2048,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/resnet50_omz.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    pretrained=None,
+    backbone=dict(
+        type='OmzBackboneCls',
+        mode='train',
+        model_path='public/resnet-50-caffe2/FP32/resnet-50-caffe2.xml',
+        last_layer_name='gpu_0/res5_2_branch2c_bn_3',
+        normalized_img_input=True
+    ),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=2048,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/_base_/vgg19_omz.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    pretrained=None,
+    backbone=dict(
+        type='OmzBackboneCls',
+        mode='train',
+        model_path='public/vgg19-caffe2/FP32/vgg19-caffe2.xml',
+        last_layer_name='conv5_4_2',
+        normalized_img_input=True
+    ),
+    neck=dict(
+        type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=512,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))

models/classification/densenet_121_omz.yaml

@@ -0,0 +1,5 @@
+_base_: _base_/densenet_121_omz.py
+
+model:
+    task: classification
+

models/classification/efficientnet_b0_omz.yaml

@@ -0,0 +1,5 @@
+_base_: _base_/efficientnet_b0_omz.py
+
+model:
+    task: classification
+

models/classification/efficientnet_b5_omz.yaml

@@ -0,0 +1,5 @@
+_base_: _base_/efficientnet_b5_omz.py
+
+model:
+    task: classification
+

models/classification/efficientnet_b7_omz.yaml

@@ -0,0 +1,5 @@
+_base_: _base_/efficientnet_b7_omz.py
+
+model:
+    task: classification
+

models/classification/googlenet_v1_omz.yaml

@@ -0,0 +1,5 @@
+_base_: _base_/googlenet_v1_omz.py
+
+model:
+    task: classification
+

models/classification/googlenet_v2_omz.yaml

@@ -0,0 +1,5 @@
+_base_: _base_/googlenet_v2_omz.py
+
+model:
+    task: classification
+

models/classification/googlenet_v3_omz.yaml

@@ -0,0 +1,5 @@
+_base_: _base_/googlenet_v3_omz.py
+
+model:
+    task: classification
+

models/classification/googlenet_v4_omz.yaml

@@ -0,0 +1,5 @@
+_base_: _base_/googlenet_v4_omz.py
+
+model:
+    task: classification
+