release code

Dassl.ProGrad.pytorch/configs/trainers/da/dael/digit5.yaml

@@ -0,0 +1,20 @@
+DATALOADER:
+  TRAIN_X:
+    SAMPLER: "RandomDomainSampler"
+    BATCH_SIZE: 256
+  TRAIN_U:
+    SAME_AS_X: False
+    BATCH_SIZE: 64
+  TEST:
+    BATCH_SIZE: 256
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.05
+  STEPSIZE: [30]
+  MAX_EPOCH: 30
+  LR_SCHEDULER: "cosine"
+
+TRAINER:
+  DAEL:
+    STRONG_TRANSFORMS: ["randaugment2", "normalize"]

Dassl.ProGrad.pytorch/configs/trainers/da/dael/domainnet.yaml

@@ -0,0 +1,20 @@
+DATALOADER:
+  NUM_WORKERS: 4
+  TRAIN_X:
+    SAMPLER: "RandomDomainSampler"
+    BATCH_SIZE: 30
+  TRAIN_U:
+    SAME_AS_X: False
+    BATCH_SIZE: 6
+  TEST:
+    BATCH_SIZE: 30
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.002
+  MAX_EPOCH: 40
+  LR_SCHEDULER: "cosine"
+
+TRAINER:
+  DAEL:
+    STRONG_TRANSFORMS: ["random_flip", "cutout", "randaugment2", "normalize"]

Dassl.ProGrad.pytorch/configs/trainers/da/dael/mini_domainnet.yaml

@@ -0,0 +1,20 @@
+DATALOADER:
+  NUM_WORKERS: 8
+  TRAIN_X:
+    SAMPLER: "RandomDomainSampler"
+    BATCH_SIZE: 192
+  TRAIN_U:
+    SAME_AS_X: False
+    BATCH_SIZE: 64
+  TEST:
+    BATCH_SIZE: 200
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.005
+  MAX_EPOCH: 60
+  LR_SCHEDULER: "cosine"
+
+TRAINER:
+  DAEL:
+    STRONG_TRANSFORMS: ["random_flip", "cutout", "randaugment2", "normalize"]

Dassl.ProGrad.pytorch/configs/trainers/da/m3sda/digit5.yaml

@@ -0,0 +1,16 @@
+DATALOADER:
+  TRAIN_X:
+    SAMPLER: "RandomDomainSampler"
+    BATCH_SIZE: 256
+  TRAIN_U:
+    SAME_AS_X: False
+    BATCH_SIZE: 64
+  TEST:
+    BATCH_SIZE: 256
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.05
+  STEPSIZE: [30]
+  MAX_EPOCH: 30
+  LR_SCHEDULER: "cosine"

Dassl.ProGrad.pytorch/configs/trainers/da/m3sda/domainnet.yaml

@@ -0,0 +1,16 @@
+DATALOADER:
+  NUM_WORKERS: 4
+  TRAIN_X:
+    SAMPLER: "RandomDomainSampler"
+    BATCH_SIZE: 30
+  TRAIN_U:
+    SAME_AS_X: False
+    BATCH_SIZE: 6
+  TEST:
+    BATCH_SIZE: 30
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.002
+  MAX_EPOCH: 40
+  LR_SCHEDULER: "cosine"

Dassl.ProGrad.pytorch/configs/trainers/da/m3sda/mini_domainnet.yaml

@@ -0,0 +1,16 @@
+DATALOADER:
+  NUM_WORKERS: 8
+  TRAIN_X:
+    SAMPLER: "RandomDomainSampler"
+    BATCH_SIZE: 192
+  TRAIN_U:
+    SAME_AS_X: False
+    BATCH_SIZE: 64
+  TEST:
+    BATCH_SIZE: 200
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.005
+  MAX_EPOCH: 60
+  LR_SCHEDULER: "cosine"

Dassl.ProGrad.pytorch/configs/trainers/da/source_only/digit5.yaml

@@ -0,0 +1,12 @@
+DATALOADER:
+  TRAIN_X:
+    BATCH_SIZE: 256
+  TEST:
+    BATCH_SIZE: 256
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.05
+  STEPSIZE: [30]
+  MAX_EPOCH: 30
+  LR_SCHEDULER: "cosine"

Dassl.ProGrad.pytorch/configs/trainers/da/source_only/mini_domainnet.yaml

@@ -0,0 +1,12 @@
+DATALOADER:
+  NUM_WORKERS: 8
+  TRAIN_X:
+    BATCH_SIZE: 128
+  TEST:
+    BATCH_SIZE: 128
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.005
+  MAX_EPOCH: 60
+  LR_SCHEDULER: "cosine"

Dassl.ProGrad.pytorch/configs/trainers/da/source_only/office31.yaml

@@ -0,0 +1,11 @@
+DATALOADER:
+  TRAIN_X:
+    BATCH_SIZE: 32
+  TEST:
+    BATCH_SIZE: 32
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.002
+  STEPSIZE: [20]
+  MAX_EPOCH: 20

Dassl.ProGrad.pytorch/configs/trainers/da/source_only/visda17.yaml

@@ -0,0 +1,15 @@
+DATALOADER:
+  TRAIN_X:
+    BATCH_SIZE: 32
+  TEST:
+    BATCH_SIZE: 32
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.0001
+  STEPSIZE: [2]
+  MAX_EPOCH: 2
+
+TRAIN:
+  PRINT_FREQ: 50
+  COUNT_ITER: "train_u"

Dassl.ProGrad.pytorch/configs/trainers/dg/dael/digits_dg.yaml

@@ -0,0 +1,16 @@
+DATALOADER:
+  TRAIN_X:
+    SAMPLER: "RandomDomainSampler"
+    BATCH_SIZE: 120
+  TEST:
+    BATCH_SIZE: 100
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.05
+  STEPSIZE: [20]
+  MAX_EPOCH: 50
+
+TRAINER:
+  DAEL:
+    STRONG_TRANSFORMS: ["randaugment2", "normalize"]

Dassl.ProGrad.pytorch/configs/trainers/dg/dael/office_home_dg.yaml

@@ -0,0 +1,16 @@
+DATALOADER:
+  TRAIN_X:
+    SAMPLER: "RandomDomainSampler"
+    BATCH_SIZE: 30
+  TEST:
+    BATCH_SIZE: 100
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.002
+  MAX_EPOCH: 40
+  LR_SCHEDULER: "cosine"
+
+TRAINER:
+  DAEL:
+    STRONG_TRANSFORMS: ["random_flip", "cutout", "randaugment2", "normalize"]

Dassl.ProGrad.pytorch/configs/trainers/dg/dael/pacs.yaml

@@ -0,0 +1,16 @@
+DATALOADER:
+  TRAIN_X:
+    SAMPLER: "RandomDomainSampler"
+    BATCH_SIZE: 30
+  TEST:
+    BATCH_SIZE: 100
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.002
+  MAX_EPOCH: 40
+  LR_SCHEDULER: "cosine"
+
+TRAINER:
+  DAEL:
+    STRONG_TRANSFORMS: ["random_flip", "cutout", "randaugment2", "normalize"]

Dassl.ProGrad.pytorch/configs/trainers/dg/ddaig/digits_dg.yaml

@@ -0,0 +1,20 @@
+INPUT:
+  PIXEL_MEAN: [0., 0., 0.]
+  PIXEL_STD: [1., 1., 1.]
+
+DATALOADER:
+  TRAIN_X:
+    BATCH_SIZE: 128
+  TEST:
+    BATCH_SIZE: 128
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.05
+  STEPSIZE: [20]
+  MAX_EPOCH: 50
+
+TRAINER:
+  DDAIG:
+    G_ARCH: "fcn_3x32_gctx"
+    LMDA: 0.3

Dassl.ProGrad.pytorch/configs/trainers/dg/ddaig/office_home_dg.yaml

@@ -0,0 +1,21 @@
+INPUT:
+  PIXEL_MEAN: [0., 0., 0.]
+  PIXEL_STD: [1., 1., 1.]
+
+DATALOADER:
+  TRAIN_X:
+    BATCH_SIZE: 16
+  TEST:
+    BATCH_SIZE: 16
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.0005
+  STEPSIZE: [20]
+  MAX_EPOCH: 25
+
+TRAINER:
+  DDAIG:
+    G_ARCH: "fcn_3x64_gctx"
+    WARMUP: 3
+    LMDA: 0.3

Dassl.ProGrad.pytorch/configs/trainers/dg/ddaig/pacs.yaml

@@ -0,0 +1,21 @@
+INPUT:
+  PIXEL_MEAN: [0., 0., 0.]
+  PIXEL_STD: [1., 1., 1.]
+
+DATALOADER:
+  TRAIN_X:
+    BATCH_SIZE: 16
+  TEST:
+    BATCH_SIZE: 16
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.0005
+  STEPSIZE: [20]
+  MAX_EPOCH: 25
+
+TRAINER:
+  DDAIG:
+    G_ARCH: "fcn_3x64_gctx"
+    WARMUP: 3
+    LMDA: 0.3

Dassl.ProGrad.pytorch/configs/trainers/dg/vanilla/digits_dg.yaml

@@ -0,0 +1,15 @@
+DATALOADER:
+  TRAIN_X:
+    BATCH_SIZE: 128
+  TEST:
+    BATCH_SIZE: 100
+  NUM_WORKERS: 8
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.05
+  STEPSIZE: [20]
+  MAX_EPOCH: 50
+
+TRAIN:
+  PRINT_FREQ: 20

Dassl.ProGrad.pytorch/configs/trainers/dg/vanilla/mini_domainnet.yaml

@@ -0,0 +1,12 @@
+DATALOADER:
+  NUM_WORKERS: 8
+  TRAIN_X:
+    BATCH_SIZE: 128
+  TEST:
+    BATCH_SIZE: 128
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.005
+  MAX_EPOCH: 60
+  LR_SCHEDULER: "cosine"

Dassl.ProGrad.pytorch/configs/trainers/dg/vanilla/office_home_dg.yaml

@@ -0,0 +1,12 @@
+DATALOADER:
+  TRAIN_X:
+    BATCH_SIZE: 64
+  TEST:
+    BATCH_SIZE: 100
+  NUM_WORKERS: 8
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.001
+  MAX_EPOCH: 50
+  LR_SCHEDULER: "cosine"

Dassl.ProGrad.pytorch/configs/trainers/dg/vanilla/pacs.yaml

@@ -0,0 +1,12 @@
+DATALOADER:
+  TRAIN_X:
+    BATCH_SIZE: 64
+  TEST:
+    BATCH_SIZE: 100
+  NUM_WORKERS: 8
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.001
+  MAX_EPOCH: 50
+  LR_SCHEDULER: "cosine"

Dassl.ProGrad.pytorch/configs/trainers/ssl/fixmatch/cifar10.yaml

@@ -0,0 +1,23 @@
+DATALOADER:
+  TRAIN_X:
+    BATCH_SIZE: 64
+  TRAIN_U:
+    SAME_AS_X: False
+    BATCH_SIZE: 448
+  TEST:
+    BATCH_SIZE: 500
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.05
+  STEPSIZE: [4000]
+  MAX_EPOCH: 4000
+  LR_SCHEDULER: "cosine"
+
+TRAIN:
+  COUNT_ITER: "train_u"
+  PRINT_FREQ: 10
+
+TRAINER:
+  FIXMATCH:
+    STRONG_TRANSFORMS: ["random_flip", "randaugment_fixmatch", "normalize", "cutout"]