Upload 164 files

BAN-vit-b16-clip-mit-b0/distill/config.py

@@ -0,0 +1,102 @@
+_base_ = [
+    '/home/liuziyuan/proj/rmcd-kd/configs/_base_/models/KD-ban_vit-b16.py', 
+    '/home/liuziyuan/proj/rmcd-kd/configs/common/standard_512x512_200k_cgwx.py']
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+checkpoint_student = None
+checkpoint_teacher_l = None
+checkpoint_teacher_m = None
+checkpoint_teacher_s = None
+
+# model settings
+model = dict(
+    init_cfg=dict(type='Pretrained', checkpoint=checkpoint_student),
+    # teacher large    
+    init_cfg_t_l = dict(type='Pretrained', checkpoint=checkpoint_teacher_l),
+    # teacher medium    
+    init_cfg_t_m = dict(type='Pretrained', checkpoint=checkpoint_teacher_m),
+    # teacher small    
+    init_cfg_t_s = dict(type='Pretrained', checkpoint=checkpoint_teacher_s),
+    
+    asymetric_input=True,
+    encoder_resolution=dict(
+        size=(224, 224),
+        mode='bilinear'),
+    image_encoder=dict(
+        frozen_exclude=[]),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            clip_channels=768,
+            fusion_index=[1, 2, 3],
+            side_enc_cfg=dict(
+                type='mmseg.MixVisionTransformer',
+                in_channels=3,
+                embed_dims=32,
+                num_stages=4,
+                num_layers=[2, 2, 2, 2],
+                num_heads=[1, 2, 5, 8],
+                patch_sizes=[7, 3, 3, 3],
+                sr_ratios=[8, 4, 2, 1],
+                out_indices=(0, 1, 2, 3),
+                mlp_ratio=4,
+                qkv_bias=True,
+                drop_rate=0.0,
+                attn_drop_rate=0.0,
+                drop_path_rate=0.1)),
+        ban_dec_cfg=dict(
+            type='BAN_MLPDecoder',
+            in_channels=[32, 64, 160, 256],
+            channels=128,
+            dropout_ratio=0.1,
+            num_classes=2,
+            norm_cfg=dict(type='SyncBN', requires_grad=True),
+            align_corners=False)),
+    test_cfg=dict(mode='slide', crop_size=crop_size, stride=(crop_size[0]//2, crop_size[1]//2)))
+
+optim_wrapper = dict(
+    _delete_=True,
+    type='AmpOptimWrapper',
+    optimizer=dict(
+        type='AdamW', lr=0.0001, betas=(0.9, 0.999), weight_decay=0.0001),
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_encoder': dict(lr_mult=0.1, decay_mult=1.0),
+            'norm': dict(decay_mult=0.),
+            'mask_decoder': dict(lr_mult=10.)
+        }),
+    loss_scale='dynamic',
+    clip_grad=dict(max_norm=0.01, norm_type=2))
+
+train_dataloader = dict(batch_size=8, num_workers=8)
+val_dataloader = dict(batch_size=1, num_workers=1)
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BAN-vit-b16-clip-mit-b0/initial/config.py

@@ -0,0 +1,92 @@
+_base_ = [
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/_base_/models/ban_vit-b16.py', 
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/common/standard_512x512_200k_cgwx.py']
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b0_20220624-7e0fe6dd.pth'  # noqa
+
+model = dict(
+    asymetric_input=True,
+    encoder_resolution=dict(
+        size=(224, 224),
+        mode='bilinear'),
+    image_encoder=dict(
+        frozen_exclude=[]),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            clip_channels=768,
+            fusion_index=[1, 2, 3],
+            side_enc_cfg=dict(
+                type='mmseg.MixVisionTransformer',
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                in_channels=3,
+                embed_dims=32,
+                num_stages=4,
+                num_layers=[2, 2, 2, 2],
+                num_heads=[1, 2, 5, 8],
+                patch_sizes=[7, 3, 3, 3],
+                sr_ratios=[8, 4, 2, 1],
+                out_indices=(0, 1, 2, 3),
+                mlp_ratio=4,
+                qkv_bias=True,
+                drop_rate=0.0,
+                attn_drop_rate=0.0,
+                drop_path_rate=0.1)),
+        ban_dec_cfg=dict(
+            type='BAN_MLPDecoder',
+            in_channels=[32, 64, 160, 256],
+            channels=128,
+            dropout_ratio=0.1,
+            num_classes=2,
+            norm_cfg=dict(type='SyncBN', requires_grad=True),
+            align_corners=False)),
+    test_cfg=dict(mode='slide', crop_size=crop_size, stride=(crop_size[0]//2, crop_size[1]//2)))
+
+optim_wrapper = dict(
+    _delete_=True,
+    type='AmpOptimWrapper',
+    optimizer=dict(
+        type='AdamW', lr=0.0001, betas=(0.9, 0.999), weight_decay=0.0001),
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_encoder': dict(lr_mult=0.1, decay_mult=1.0),
+            'norm': dict(decay_mult=0.),
+            'mask_decoder': dict(lr_mult=10.)
+        }),
+    loss_scale='dynamic',
+    clip_grad=dict(max_norm=0.01, norm_type=2))
+
+train_dataloader = dict(batch_size=8, num_workers=8)
+val_dataloader = dict(batch_size=1, num_workers=1)
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BAN-vit-b16-clip-mit-b0/large/config.py

@@ -0,0 +1,89 @@
+_base_ = [
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/_base_/models/ban_vit-b16.py', 
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/common/train_large_512x512_100k_cgwx.py']
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b0_20220624-7e0fe6dd.pth'  # noqa
+
+model = dict(
+    asymetric_input=True,
+    encoder_resolution=dict(
+        size=(224, 224),
+        mode='bilinear'),
+    image_encoder=dict(
+        frozen_exclude=[]),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            clip_channels=768,
+            fusion_index=[1, 2, 3],
+            side_enc_cfg=dict(
+                type='mmseg.MixVisionTransformer',
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                in_channels=3,
+                embed_dims=32,
+                num_stages=4,
+                num_layers=[2, 2, 2, 2],
+                num_heads=[1, 2, 5, 8],
+                patch_sizes=[7, 3, 3, 3],
+                sr_ratios=[8, 4, 2, 1],
+                out_indices=(0, 1, 2, 3),
+                mlp_ratio=4,
+                qkv_bias=True,
+                drop_rate=0.0,
+                attn_drop_rate=0.0,
+                drop_path_rate=0.1)),
+        ban_dec_cfg=dict(
+            type='BAN_MLPDecoder',
+            in_channels=[32, 64, 160, 256],
+            channels=128,
+            dropout_ratio=0.1,
+            num_classes=2,
+            norm_cfg=dict(type='SyncBN', requires_grad=True),
+            align_corners=False)),
+    test_cfg=dict(mode='slide', crop_size=crop_size, stride=(crop_size[0]//2, crop_size[1]//2)))
+
+optim_wrapper = dict(
+    _delete_=True,
+    type='AmpOptimWrapper',
+    optimizer=dict(
+        type='AdamW', lr=0.0001, betas=(0.9, 0.999), weight_decay=0.0001),
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_encoder': dict(lr_mult=0.1, decay_mult=1.0),
+            'norm': dict(decay_mult=0.),
+            'mask_decoder': dict(lr_mult=10.)
+        }),
+    loss_scale='dynamic',
+    clip_grad=dict(max_norm=0.01, norm_type=2))
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BAN-vit-b16-clip-mit-b0/medium/config.py

@@ -0,0 +1,89 @@
+_base_ = [
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/_base_/models/ban_vit-b16.py', 
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/common/train_medium_512x512_100k_cgwx.py']
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b0_20220624-7e0fe6dd.pth'  # noqa
+
+model = dict(
+    asymetric_input=True,
+    encoder_resolution=dict(
+        size=(224, 224),
+        mode='bilinear'),
+    image_encoder=dict(
+        frozen_exclude=[]),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            clip_channels=768,
+            fusion_index=[1, 2, 3],
+            side_enc_cfg=dict(
+                type='mmseg.MixVisionTransformer',
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                in_channels=3,
+                embed_dims=32,
+                num_stages=4,
+                num_layers=[2, 2, 2, 2],
+                num_heads=[1, 2, 5, 8],
+                patch_sizes=[7, 3, 3, 3],
+                sr_ratios=[8, 4, 2, 1],
+                out_indices=(0, 1, 2, 3),
+                mlp_ratio=4,
+                qkv_bias=True,
+                drop_rate=0.0,
+                attn_drop_rate=0.0,
+                drop_path_rate=0.1)),
+        ban_dec_cfg=dict(
+            type='BAN_MLPDecoder',
+            in_channels=[32, 64, 160, 256],
+            channels=128,
+            dropout_ratio=0.1,
+            num_classes=2,
+            norm_cfg=dict(type='SyncBN', requires_grad=True),
+            align_corners=False)),
+    test_cfg=dict(mode='slide', crop_size=crop_size, stride=(crop_size[0]//2, crop_size[1]//2)))
+
+optim_wrapper = dict(
+    _delete_=True,
+    type='AmpOptimWrapper',
+    optimizer=dict(
+        type='AdamW', lr=0.0001, betas=(0.9, 0.999), weight_decay=0.0001),
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_encoder': dict(lr_mult=0.1, decay_mult=1.0),
+            'norm': dict(decay_mult=0.),
+            'mask_decoder': dict(lr_mult=10.)
+        }),
+    loss_scale='dynamic',
+    clip_grad=dict(max_norm=0.01, norm_type=2))
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BAN-vit-b16-clip-mit-b0/small/config.py

@@ -0,0 +1,89 @@
+_base_ = [
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/_base_/models/ban_vit-b16.py', 
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/common/train_small_512x512_100k_cgwx.py']
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b0_20220624-7e0fe6dd.pth'  # noqa
+
+model = dict(
+    asymetric_input=True,
+    encoder_resolution=dict(
+        size=(224, 224),
+        mode='bilinear'),
+    image_encoder=dict(
+        frozen_exclude=[]),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            clip_channels=768,
+            fusion_index=[1, 2, 3],
+            side_enc_cfg=dict(
+                type='mmseg.MixVisionTransformer',
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                in_channels=3,
+                embed_dims=32,
+                num_stages=4,
+                num_layers=[2, 2, 2, 2],
+                num_heads=[1, 2, 5, 8],
+                patch_sizes=[7, 3, 3, 3],
+                sr_ratios=[8, 4, 2, 1],
+                out_indices=(0, 1, 2, 3),
+                mlp_ratio=4,
+                qkv_bias=True,
+                drop_rate=0.0,
+                attn_drop_rate=0.0,
+                drop_path_rate=0.1)),
+        ban_dec_cfg=dict(
+            type='BAN_MLPDecoder',
+            in_channels=[32, 64, 160, 256],
+            channels=128,
+            dropout_ratio=0.1,
+            num_classes=2,
+            norm_cfg=dict(type='SyncBN', requires_grad=True),
+            align_corners=False)),
+    test_cfg=dict(mode='slide', crop_size=crop_size, stride=(crop_size[0]//2, crop_size[1]//2)))
+
+optim_wrapper = dict(
+    _delete_=True,
+    type='AmpOptimWrapper',
+    optimizer=dict(
+        type='AdamW', lr=0.0001, betas=(0.9, 0.999), weight_decay=0.0001),
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_encoder': dict(lr_mult=0.1, decay_mult=1.0),
+            'norm': dict(decay_mult=0.),
+            'mask_decoder': dict(lr_mult=10.)
+        }),
+    loss_scale='dynamic',
+    clip_grad=dict(max_norm=0.01, norm_type=2))
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BAN-vit-b16-clip-mit-b2/initial/best_mIoU_iter_15000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:184e2e7aadcb8e6af5ad71fc3dc2404d482ef5eaecf981b5343f8ea274bd0376
+size 367861071

BAN-vit-b16-clip-mit-b2/initial/config.py

@@ -0,0 +1,15 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-clip-mit-b0/initial/config.py']
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model = dict(
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))

BAN-vit-b16-clip-mit-b2/large/best_mIoU_iter_1000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14580d22a98d70d35a75cbcdbbca2f476b2cda63de66a0d1fb4add85d558b138
+size 365786767

BAN-vit-b16-clip-mit-b2/large/config.py

@@ -0,0 +1,15 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-clip-mit-b0/large/config.py']
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model = dict(
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))

BAN-vit-b16-clip-mit-b2/medium/best_mIoU_iter_41000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:133dcc00fa7b44c63f3030298839dbe1ca38077997a90f5886ed119fe9a0c69f
+size 371713615

BAN-vit-b16-clip-mit-b2/medium/config.py

@@ -0,0 +1,15 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-clip-mit-b0/medium/config.py']
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model = dict(
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))

BAN-vit-b16-clip-mit-b2/small/best_mIoU_iter_23000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93578f2d4d362516e8842e67760265f2639a8464d8673d0859ee1ace769d449e
+size 369088591

BAN-vit-b16-clip-mit-b2/small/config.py

@@ -0,0 +1,15 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-clip-mit-b0/small/config.py']
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model = dict(
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))

BAN-vit-b16-in21k-mit-b0/initial/config.py

@@ -0,0 +1,92 @@
+_base_ = [
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/_base_/models/ban_vit-b16.py', 
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/common/standard_512x512_200k_cgwx.py']
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+vit_checkpoint_file = 'pretrain/augreg_B_16-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.0-sd_0.0.pth'  # noqa
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b0_20220624-7e0fe6dd.pth'  # noqa
+
+model = dict(
+    pretrained=None,
+    asymetric_input=True,
+    encoder_resolution=dict(
+        size=(224, 224),
+        mode='bilinear'),
+    image_encoder=dict(
+        type='mmseg.VisionTransformer',
+        init_cfg=dict(
+            type='Pretrained', checkpoint=vit_checkpoint_file),
+        img_size=(224, 224),
+        patch_size=16,
+        in_channels=3,
+        embed_dims=768,
+        num_layers=12,
+        num_heads=12,
+        mlp_ratio=4,
+        out_indices=(5, 8, 11),
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.0,
+        with_cls_token=True,
+        norm_cfg=dict(type='LN', eps=1e-6),
+        act_cfg=dict(type='GELU'),
+        norm_eval=False,
+        patch_bias=True,
+        interpolate_mode='bicubic',
+        frozen_exclude=[]),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            clip_channels=768,
+            fusion_index=[1, 2, 3],
+            side_enc_cfg=dict(
+                type='mmseg.MixVisionTransformer',
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                in_channels=3,
+                embed_dims=32,
+                num_stages=4,
+                num_layers=[2, 2, 2, 2],
+                num_heads=[1, 2, 5, 8],
+                patch_sizes=[7, 3, 3, 3],
+                sr_ratios=[8, 4, 2, 1],
+                out_indices=(0, 1, 2, 3),
+                mlp_ratio=4,
+                qkv_bias=True,
+                drop_rate=0.0,
+                attn_drop_rate=0.0,
+                drop_path_rate=0.1)),
+        ban_dec_cfg=dict(
+            type='BAN_MLPDecoder',
+            in_channels=[32, 64, 160, 256],
+            channels=128,
+            dropout_ratio=0.1,
+            num_classes=2,
+            norm_cfg=dict(type='SyncBN', requires_grad=True),
+            align_corners=False,
+        )),
+    test_cfg=dict(mode='slide', crop_size=crop_size, stride=(crop_size[0]//2, crop_size[1]//2)))
+
+optim_wrapper = dict(
+    _delete_=True,
+    type='AmpOptimWrapper',
+    optimizer=dict(
+        type='AdamW', lr=0.0001, betas=(0.9, 0.999), weight_decay=0.0001),
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_encoder': dict(lr_mult=0.1, decay_mult=1.0),
+            'norm': dict(decay_mult=0.),
+            'mask_decoder': dict(lr_mult=10.)
+        }),
+    loss_scale='dynamic',
+    clip_grad=dict(max_norm=0.01, norm_type=2))
+
+train_dataloader = dict(batch_size=8, num_workers=8)
+val_dataloader = dict(batch_size=1, num_workers=1)
+
+# find_unused_parameters=True

BAN-vit-b16-in21k-mit-b0/large/config.py

@@ -0,0 +1,113 @@
+_base_ = [
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/_base_/models/ban_vit-b16.py', 
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/common/train_large_512x512_100k_cgwx.py']
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+vit_checkpoint_file = 'pretrain/augreg_B_16-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.0-sd_0.0.pth'  # noqa
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b0_20220624-7e0fe6dd.pth'  # noqa
+
+model = dict(
+    pretrained=None,
+    asymetric_input=True,
+    encoder_resolution=dict(
+        size=(224, 224),
+        mode='bilinear'),
+    image_encoder=dict(
+        type='mmseg.VisionTransformer',
+        init_cfg=dict(
+            type='Pretrained', checkpoint=vit_checkpoint_file),
+        img_size=(224, 224),
+        patch_size=16,
+        in_channels=3,
+        embed_dims=768,
+        num_layers=12,
+        num_heads=12,
+        mlp_ratio=4,
+        out_indices=(5, 8, 11),
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.0,
+        with_cls_token=True,
+        norm_cfg=dict(type='LN', eps=1e-6),
+        act_cfg=dict(type='GELU'),
+        norm_eval=False,
+        patch_bias=True,
+        interpolate_mode='bicubic',
+        frozen_exclude=[]),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            clip_channels=768,
+            fusion_index=[1, 2, 3],
+            side_enc_cfg=dict(
+                type='mmseg.MixVisionTransformer',
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                in_channels=3,
+                embed_dims=32,
+                num_stages=4,
+                num_layers=[2, 2, 2, 2],
+                num_heads=[1, 2, 5, 8],
+                patch_sizes=[7, 3, 3, 3],
+                sr_ratios=[8, 4, 2, 1],
+                out_indices=(0, 1, 2, 3),
+                mlp_ratio=4,
+                qkv_bias=True,
+                drop_rate=0.0,
+                attn_drop_rate=0.0,
+                drop_path_rate=0.1)),
+        ban_dec_cfg=dict(
+            type='BAN_MLPDecoder',
+            in_channels=[32, 64, 160, 256],
+            channels=128,
+            dropout_ratio=0.1,
+            num_classes=2,
+            norm_cfg=dict(type='SyncBN', requires_grad=True),
+            align_corners=False,
+        )),
+    test_cfg=dict(mode='slide', crop_size=crop_size, stride=(crop_size[0]//2, crop_size[1]//2)))
+
+optim_wrapper = dict(
+    _delete_=True,
+    type='AmpOptimWrapper',
+    optimizer=dict(
+        type='AdamW', lr=0.0001, betas=(0.9, 0.999), weight_decay=0.0001),
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_encoder': dict(lr_mult=0.1, decay_mult=1.0),
+            'norm': dict(decay_mult=0.),
+            'mask_decoder': dict(lr_mult=10.)
+        }),
+    loss_scale='dynamic',
+    clip_grad=dict(max_norm=0.01, norm_type=2))
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BAN-vit-b16-in21k-mit-b0/medium/config.py

@@ -0,0 +1,113 @@
+_base_ = [
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/_base_/models/ban_vit-b16.py', 
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/common/train_medium_512x512_100k_cgwx.py']
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+vit_checkpoint_file = 'pretrain/augreg_B_16-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.0-sd_0.0.pth'  # noqa
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b0_20220624-7e0fe6dd.pth'  # noqa
+
+model = dict(
+    pretrained=None,
+    asymetric_input=True,
+    encoder_resolution=dict(
+        size=(224, 224),
+        mode='bilinear'),
+    image_encoder=dict(
+        type='mmseg.VisionTransformer',
+        init_cfg=dict(
+            type='Pretrained', checkpoint=vit_checkpoint_file),
+        img_size=(224, 224),
+        patch_size=16,
+        in_channels=3,
+        embed_dims=768,
+        num_layers=12,
+        num_heads=12,
+        mlp_ratio=4,
+        out_indices=(5, 8, 11),
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.0,
+        with_cls_token=True,
+        norm_cfg=dict(type='LN', eps=1e-6),
+        act_cfg=dict(type='GELU'),
+        norm_eval=False,
+        patch_bias=True,
+        interpolate_mode='bicubic',
+        frozen_exclude=[]),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            clip_channels=768,
+            fusion_index=[1, 2, 3],
+            side_enc_cfg=dict(
+                type='mmseg.MixVisionTransformer',
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                in_channels=3,
+                embed_dims=32,
+                num_stages=4,
+                num_layers=[2, 2, 2, 2],
+                num_heads=[1, 2, 5, 8],
+                patch_sizes=[7, 3, 3, 3],
+                sr_ratios=[8, 4, 2, 1],
+                out_indices=(0, 1, 2, 3),
+                mlp_ratio=4,
+                qkv_bias=True,
+                drop_rate=0.0,
+                attn_drop_rate=0.0,
+                drop_path_rate=0.1)),
+        ban_dec_cfg=dict(
+            type='BAN_MLPDecoder',
+            in_channels=[32, 64, 160, 256],
+            channels=128,
+            dropout_ratio=0.1,
+            num_classes=2,
+            norm_cfg=dict(type='SyncBN', requires_grad=True),
+            align_corners=False,
+        )),
+    test_cfg=dict(mode='slide', crop_size=crop_size, stride=(crop_size[0]//2, crop_size[1]//2)))
+
+optim_wrapper = dict(
+    _delete_=True,
+    type='AmpOptimWrapper',
+    optimizer=dict(
+        type='AdamW', lr=0.0001, betas=(0.9, 0.999), weight_decay=0.0001),
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_encoder': dict(lr_mult=0.1, decay_mult=1.0),
+            'norm': dict(decay_mult=0.),
+            'mask_decoder': dict(lr_mult=10.)
+        }),
+    loss_scale='dynamic',
+    clip_grad=dict(max_norm=0.01, norm_type=2))
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BAN-vit-b16-in21k-mit-b0/small/config.py

@@ -0,0 +1,113 @@
+_base_ = [
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/_base_/models/ban_vit-b16.py', 
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/common/train_small_512x512_100k_cgwx.py']
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+vit_checkpoint_file = 'pretrain/augreg_B_16-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.0-sd_0.0.pth'  # noqa
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b0_20220624-7e0fe6dd.pth'  # noqa
+
+model = dict(
+    pretrained=None,
+    asymetric_input=True,
+    encoder_resolution=dict(
+        size=(224, 224),
+        mode='bilinear'),
+    image_encoder=dict(
+        type='mmseg.VisionTransformer',
+        init_cfg=dict(
+            type='Pretrained', checkpoint=vit_checkpoint_file),
+        img_size=(224, 224),
+        patch_size=16,
+        in_channels=3,
+        embed_dims=768,
+        num_layers=12,
+        num_heads=12,
+        mlp_ratio=4,
+        out_indices=(5, 8, 11),
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.0,
+        with_cls_token=True,
+        norm_cfg=dict(type='LN', eps=1e-6),
+        act_cfg=dict(type='GELU'),
+        norm_eval=False,
+        patch_bias=True,
+        interpolate_mode='bicubic',
+        frozen_exclude=[]),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            clip_channels=768,
+            fusion_index=[1, 2, 3],
+            side_enc_cfg=dict(
+                type='mmseg.MixVisionTransformer',
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                in_channels=3,
+                embed_dims=32,
+                num_stages=4,
+                num_layers=[2, 2, 2, 2],
+                num_heads=[1, 2, 5, 8],
+                patch_sizes=[7, 3, 3, 3],
+                sr_ratios=[8, 4, 2, 1],
+                out_indices=(0, 1, 2, 3),
+                mlp_ratio=4,
+                qkv_bias=True,
+                drop_rate=0.0,
+                attn_drop_rate=0.0,
+                drop_path_rate=0.1)),
+        ban_dec_cfg=dict(
+            type='BAN_MLPDecoder',
+            in_channels=[32, 64, 160, 256],
+            channels=128,
+            dropout_ratio=0.1,
+            num_classes=2,
+            norm_cfg=dict(type='SyncBN', requires_grad=True),
+            align_corners=False,
+        )),
+    test_cfg=dict(mode='slide', crop_size=crop_size, stride=(crop_size[0]//2, crop_size[1]//2)))
+
+optim_wrapper = dict(
+    _delete_=True,
+    type='AmpOptimWrapper',
+    optimizer=dict(
+        type='AdamW', lr=0.0001, betas=(0.9, 0.999), weight_decay=0.0001),
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_encoder': dict(lr_mult=0.1, decay_mult=1.0),
+            'norm': dict(decay_mult=0.),
+            'mask_decoder': dict(lr_mult=10.)
+        }),
+    loss_scale='dynamic',
+    clip_grad=dict(max_norm=0.01, norm_type=2))
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=200000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=200000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BAN-vit-b16-in21k-mit-b2/initial/best_mIoU_iter_30000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e408bf4ee3dab555f143f1dd975a5e643413fb7e6c3b7aa333b88cf824d0dd1e
+size 455144133

BAN-vit-b16-in21k-mit-b2/initial/config.py

@@ -0,0 +1,15 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-in21k-mit-b0/initial/config.py']
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model = dict(
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))

BAN-vit-b16-in21k-mit-b2/large/best_mIoU_iter_7000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d8fc2343b5b9c252c86b7032711da60404eff6ab2b1f575f69a29615525c46c
+size 451744005

BAN-vit-b16-in21k-mit-b2/large/config.py

@@ -0,0 +1,15 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-in21k-mit-b0/large/config.py']
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model = dict(
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))

BAN-vit-b16-in21k-mit-b2/medium/best_mIoU_iter_6000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7917af7d31f1ee68ff7fda449bc0d048b2232bf147d15fd5aca12a5a9523da6
+size 451598277

BAN-vit-b16-in21k-mit-b2/medium/config.py

@@ -0,0 +1,15 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-in21k-mit-b0/medium/config.py']
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model = dict(
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))

BAN-vit-b16-in21k-mit-b2/small/best_mIoU_iter_19000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1642ec06e0bcdced0e04a56b9dc5ec75a37f5251c1b8512045311da40a42aaa1
+size 453552645

BAN-vit-b16-in21k-mit-b2/small/config.py

@@ -0,0 +1,15 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-in21k-mit-b0/small/config.py']
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model = dict(
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))

BAN-vit-l14-clip-mit-b0/distill/config.py

@@ -0,0 +1,40 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-clip-mit-b0/distill/config.py']
+
+checkpoint_student = None
+checkpoint_teacher_l = None
+checkpoint_teacher_m = None
+checkpoint_teacher_s = None
+
+# model settings
+model = dict(
+    init_cfg=dict(type='Pretrained', checkpoint=checkpoint_student),
+    # teacher large    
+    init_cfg_t_l = dict(type='Pretrained', checkpoint=checkpoint_teacher_l),
+    # teacher medium    
+    init_cfg_t_m = dict(type='Pretrained', checkpoint=checkpoint_teacher_m),
+    # teacher small    
+    init_cfg_t_s = dict(type='Pretrained', checkpoint=checkpoint_teacher_s),
+    
+    type='DistillBAN',
+    encoder_resolution=dict(
+        size=(336, 336),
+        mode='bilinear'),
+    image_encoder=dict(
+        type='mmseg.VisionTransformer',
+        img_size=(336, 336),
+        patch_size=14,
+        patch_pad=0,
+        embed_dims=1024,
+        num_layers=18,
+        num_heads=16,
+        out_indices=(5, 11, 17)),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            fusion_index=[1, 2, 3],
+            clip_channels=1024),
+        ban_dec_cfg=dict(
+            in_channels=[32, 64, 160, 256])))
+
+train_dataloader = dict(batch_size=8, num_workers=8)
+val_dataloader = dict(batch_size=1, num_workers=1)

BAN-vit-l14-clip-mit-b0/initial/config.py

@@ -0,0 +1,29 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-clip-mit-b0/initial/config.py']
+
+pretrained = 'pretrain/clip_vit-large-patch14-336_3rdparty-0b5df9cb.pth'  # noqa
+
+model = dict(
+    type='BAN',
+    pretrained=pretrained,
+    encoder_resolution=dict(
+        size=(336, 336),
+        mode='bilinear'),
+    image_encoder=dict(
+        type='mmseg.VisionTransformer',
+        img_size=(336, 336),
+        patch_size=14,
+        patch_pad=0,
+        embed_dims=1024,
+        num_layers=18,
+        num_heads=16,
+        out_indices=(5, 11, 17)),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            fusion_index=[1, 2, 3],
+            clip_channels=1024),
+        ban_dec_cfg=dict(
+            in_channels=[32, 64, 160, 256])))
+
+train_dataloader = dict(batch_size=8, num_workers=8)
+val_dataloader = dict(batch_size=1, num_workers=1)

BAN-vit-l14-clip-mit-b0/large/config.py

@@ -0,0 +1,57 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-clip-mit-b0/large/config.py']
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+pretrained = 'pretrain/clip_vit-large-patch14-336_3rdparty-0b5df9cb.pth'  # noqa
+
+model = dict(
+    type='BAN',
+    pretrained=pretrained,
+    encoder_resolution=dict(
+        size=(336, 336),
+        mode='bilinear'),
+    image_encoder=dict(
+        type='mmseg.VisionTransformer',
+        img_size=(336, 336),
+        patch_size=14,
+        patch_pad=0,
+        embed_dims=1024,
+        num_layers=18,
+        num_heads=16,
+        out_indices=(5, 11, 17)),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            fusion_index=[1, 2, 3],
+            clip_channels=1024),
+        ban_dec_cfg=dict(
+            in_channels=[32, 64, 160, 256])))
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BAN-vit-l14-clip-mit-b0/medium/config.py

@@ -0,0 +1,57 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-clip-mit-b0/medium/config.py']
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+pretrained = 'pretrain/clip_vit-large-patch14-336_3rdparty-0b5df9cb.pth'  # noqa
+
+model = dict(
+    type='BAN',
+    pretrained=pretrained,
+    encoder_resolution=dict(
+        size=(336, 336),
+        mode='bilinear'),
+    image_encoder=dict(
+        type='mmseg.VisionTransformer',
+        img_size=(336, 336),
+        patch_size=14,
+        patch_pad=0,
+        embed_dims=1024,
+        num_layers=18,
+        num_heads=16,
+        out_indices=(5, 11, 17)),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            fusion_index=[1, 2, 3],
+            clip_channels=1024),
+        ban_dec_cfg=dict(
+            in_channels=[32, 64, 160, 256])))
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BAN-vit-l14-clip-mit-b0/small/config.py

@@ -0,0 +1,57 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-b16-clip-mit-b0/small/config.py']
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+pretrained = 'pretrain/clip_vit-large-patch14-336_3rdparty-0b5df9cb.pth'  # noqa
+
+model = dict(
+    type='BAN',
+    pretrained=pretrained,
+    encoder_resolution=dict(
+        size=(336, 336),
+        mode='bilinear'),
+    image_encoder=dict(
+        type='mmseg.VisionTransformer',
+        img_size=(336, 336),
+        patch_size=14,
+        patch_pad=0,
+        embed_dims=1024,
+        num_layers=18,
+        num_heads=16,
+        out_indices=(5, 11, 17)),
+    decode_head=dict(
+        type='BitemporalAdapterHead',
+        ban_cfg=dict(
+            fusion_index=[1, 2, 3],
+            clip_channels=1024),
+        ban_dec_cfg=dict(
+            in_channels=[32, 64, 160, 256])))
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BAN-vit-l14-clip-mit-b2/distill/best_mIoU_iter_15000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f7b3f408dfe0131948e00ef1ccbb8bab424eea22a0cf3b79c490b6cbc31d999
+size 4300572453

BAN-vit-l14-clip-mit-b2/distill/config.py

@@ -0,0 +1,27 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-l14-clip-mit-b0/distill/config.py']
+
+checkpoint_student = '/nas/datasets/lzy/RS-ChangeDetection/checkpoints_distill/BAN/vit-l14-clip-mit-b2/initial/best_mIoU_iter_17000.pth'
+checkpoint_teacher_l = '/nas/datasets/lzy/RS-ChangeDetection/checkpoints_distill/BAN/vit-l14-clip-mit-b2/large/best_mIoU_iter_24000.pth'
+checkpoint_teacher_m = '/nas/datasets/lzy/RS-ChangeDetection/checkpoints_distill/BAN/vit-l14-clip-mit-b2/medium/best_mIoU_iter_10000.pth'
+checkpoint_teacher_s = '/nas/datasets/lzy/RS-ChangeDetection/checkpoints_distill/BAN/vit-l14-clip-mit-b2/small/best_mIoU_iter_35000.pth'
+
+# model settings
+model = dict(
+    init_cfg=dict(type='Pretrained', checkpoint=checkpoint_student),
+    # teacher large    
+    init_cfg_t_l = dict(type='Pretrained', checkpoint=checkpoint_teacher_l),
+    # teacher medium    
+    init_cfg_t_m = dict(type='Pretrained', checkpoint=checkpoint_teacher_m),
+    # teacher small    
+    init_cfg_t_s = dict(type='Pretrained', checkpoint=checkpoint_teacher_s),
+    
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))
+
+train_dataloader = dict(batch_size=8, num_workers=8)
+val_dataloader = dict(batch_size=1, num_workers=1)

BAN-vit-l14-clip-mit-b2/initial/best_mIoU_iter_17000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23d8a1374f3e46e16b79dc24e6aa96264805d185c827bc42a262a8724ca3b805
+size 1022820983

BAN-vit-l14-clip-mit-b2/initial/config.py

@@ -0,0 +1,18 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-l14-clip-mit-b0/initial/config.py']
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model = dict(
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))
+
+train_dataloader = dict(batch_size=8, num_workers=8)
+val_dataloader = dict(batch_size=1, num_workers=1)

BAN-vit-l14-clip-mit-b2/large/best_mIoU_iter_24000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b08b38a473bbd18da541adfd8a0ed305365b5fe40d017a8e9cf380a26ac875e1
+size 1023807991

BAN-vit-l14-clip-mit-b2/large/config.py

@@ -0,0 +1,18 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-l14-clip-mit-b0/large/config.py']
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model = dict(
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))
+
+train_dataloader = dict(batch_size=8, num_workers=8)
+val_dataloader = dict(batch_size=1, num_workers=1)

BAN-vit-l14-clip-mit-b2/medium/best_mIoU_iter_10000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c4f2c30db1ed4e5347f9b041e627d533529b97f6743154dd8bce205f7ac1ac24
+size 1021734263

BAN-vit-l14-clip-mit-b2/medium/config.py

@@ -0,0 +1,18 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-l14-clip-mit-b0/medium/config.py']
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model = dict(
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))
+
+train_dataloader = dict(batch_size=8, num_workers=8)
+val_dataloader = dict(batch_size=1, num_workers=1)

BAN-vit-l14-clip-mit-b2/small/best_mIoU_iter_35000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d0a0d046231ad98ec3ef6a1a066f41730b8e5ba79d37355a93d3d59639177d0
+size 1025483063

BAN-vit-l14-clip-mit-b2/small/config.py

@@ -0,0 +1,18 @@
+_base_ = ['/nas/datasets/lzy/RS-ChangeDetection/Best_ckpt-KD/BAN-vit-l14-clip-mit-b0/small/config.py']
+
+checkpoint = 'https://download.openmmlab.com/mmsegmentation/v0.5/pretrain/segformer/mit_b2_20220624-66e8bf70.pth'  # noqa
+
+# model settings
+model = dict(
+    decode_head=dict(
+        ban_cfg=dict(
+            side_enc_cfg=dict(
+                init_cfg=dict(
+                    type='Pretrained', checkpoint=checkpoint),
+                embed_dims=64,
+                num_layers=[3, 4, 6, 3])),
+        ban_dec_cfg=dict(
+            in_channels=[64, 128, 320, 512])))
+
+train_dataloader = dict(batch_size=8, num_workers=8)
+val_dataloader = dict(batch_size=1, num_workers=1)

BIT/distill/best_mIoU_iter_71000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd3bb9b5333aeeff68c4df27ab52f8f25f3548b09afa715b9a204cbf458c7110
+size 57559345

BIT/distill/config.py

@@ -0,0 +1,52 @@
+_base_ = [
+    '/home/liuziyuan/proj/rmcd-kd/configs/_base_/models/KD-bit_r18.py', 
+    '/home/liuziyuan/proj/rmcd-kd/configs/common/standard_512x512_200k_cgwx.py']
+
+
+dataset_type = 'LEVIR_CD_Dataset'
+data_root = '/nas/datasets/lzy/RS-ChangeDetection/CGWX'
+
+crop_size = (512, 512)
+
+checkpoint_student = '/nas/datasets/lzy/RS-ChangeDetection/checkpoints_distill/BIT/teacher_ckpt/initial/best_mIoU_iter_80000.pth'
+checkpoint_teacher_l = '/nas/datasets/lzy/RS-ChangeDetection/checkpoints_distill/BIT/teacher_ckpt/large/best_mIoU_iter_24000.pth'
+checkpoint_teacher_m = '/nas/datasets/lzy/RS-ChangeDetection/checkpoints_distill/BIT/teacher_ckpt/medium/best_mIoU_iter_8000.pth'
+checkpoint_teacher_s = '/nas/datasets/lzy/RS-ChangeDetection/checkpoints_distill/BIT/teacher_ckpt/small/best_mIoU_iter_74000.pth'
+
+model = dict(
+    # student
+    init_cfg=dict(type='Pretrained', checkpoint=checkpoint_student),
+    # teacher large    
+    init_cfg_t_l = dict(type='Pretrained', checkpoint=checkpoint_teacher_l),
+    # teacher medium    
+    init_cfg_t_m = dict(type='Pretrained', checkpoint=checkpoint_teacher_m),
+    # teacher small    
+    init_cfg_t_s = dict(type='Pretrained', checkpoint=checkpoint_teacher_s)
+)
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BIT/initial/best_mIoU_iter_80000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:82f1f885af047d77cf07545897cae644585949239a3a5b44e66e89da9a0f545d
+size 20862109

BIT/initial/config.py

@@ -0,0 +1,3 @@
+_base_ = [
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/_base_/models/bit_r18.py', 
+    '/home/liuziyuan/proj/rmcd-kd/configs/common/standard_512x512_200k_cgwx.py']

BIT/large/best_mIoU_iter_24000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f36287b98dc665e93b015ec230aa64b43ff957eb87cc3ea145afb2a8afafc5bf
+size 14719581

BIT/large/config.py

@@ -0,0 +1,29 @@
+_base_ = [
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/_base_/models/bit_r18.py', 
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/common/train_large_512x512_100k_cgwx.py']
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BIT/medium/best_mIoU_iter_8000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b2aca7ff4a2f2f2e6ea8d1c5419f5c1fd1cc92521466a0ffd88f58625c301f0
+size 12964765

BIT/medium/config.py

@@ -0,0 +1,29 @@
+_base_ = [
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/_base_/models/bit_r18.py', 
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/common/train_medium_512x512_100k_cgwx.py']
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))

BIT/small/best_mIoU_iter_74000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cba07cd8ddb7fc840f90b2318a0f695fa256b2782c92aa615f22815dc5174b8d
+size 20356061

BIT/small/config.py

@@ -0,0 +1,29 @@
+_base_ = [
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/_base_/models/bit_r18.py', 
+    '/home/liuziyuan/proj/RS-CD/rs-cd-cgwx/configs/common/train_small_512x512_100k_cgwx.py']
+
+param_scheduler = [
+    dict(
+        type='LinearLR', start_factor=1e-6, by_epoch=False, begin=0, end=1000),
+    dict(
+        type='PolyLR',
+        power=1.0,
+        begin=1000,
+        end=100000,
+        eta_min=0.0,
+        by_epoch=False,
+    )
+]
+# training schedule for 100k
+train_cfg = dict(type='IterBasedTrainLoop', max_iters=100000, val_interval=1000)
+val_cfg = dict(type='ValLoop')
+test_cfg = dict(type='TestLoop')
+default_hooks = dict(
+    timer=dict(type='IterTimerHook'),
+    logger=dict(type='LoggerHook', interval=100, log_metric_by_epoch=False),
+    param_scheduler=dict(type='ParamSchedulerHook'),
+    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=1000,
+                    save_best='mIoU'),
+    sampler_seed=dict(type='DistSamplerSeedHook'),
+    visualization=dict(type='CDVisualizationHook', interval=1, 
+                       img_shape=(512, 512, 3)))