Upload to Main

configs/trainers/MaPLe/glib_t.yaml

@@ -0,0 +1,53 @@
+DATASET:
+  SELECTION_BATCH_SIZE: 50
+  SUBSAMPLE_CLASSES: all
+
+
+DATALOADER:
+  RETURN_IMG0: true
+  TRAIN_X:
+    BATCH_SIZE: 16
+  TEST:
+    BATCH_SIZE: 64
+  NUM_WORKERS: 2
+
+INPUT:
+  SIZE: (224, 224)
+  INTERPOLATION: "bicubic"
+  PIXEL_MEAN: [0.48145466, 0.4578275, 0.40821073]
+  PIXEL_STD: [0.26862954, 0.26130258, 0.27577711]
+  TRANSFORMS: ["random_resized_crop", "random_flip", "normalize"]
+#  CUTOUT_N: 1
+#  CUTOUT_LEN: 128
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.0026 #0.0035     0.0026 for crossdata
+  MAX_EPOCH: 5
+  LR_SCHEDULER: "cosine"
+  WARMUP_EPOCH: 1
+  WARMUP_TYPE: "constant"
+  WARMUP_CONS_LR: 1e-5
+
+TRAIN:
+  PRINT_FREQ: 20
+  CHECKPOINT_FREQ: 1
+
+MODEL:
+  BACKBONE:
+    NAME: "ViT-B/16"
+
+
+
+TEST:
+  PER_CLASS_RESULT: false
+  FINAL_MODEL: "best_val"
+
+
+
+TRAINER:
+  MAPLEG:
+    N_CTX: 4
+    CTX_INIT: "a photo of a"
+    PREC: "fp16"
+    PROMPT_DEPTH: 9

configs/trainers/MaPLe/vit_b16_base.yaml

@@ -0,0 +1,52 @@
+DATASET:
+  SELECTION_BATCH_SIZE: 50
+  SUBSAMPLE_CLASSES: base
+
+
+DATALOADER:
+  RETURN_IMG0: true
+  TRAIN_X:
+    BATCH_SIZE: 1
+  TEST:
+    BATCH_SIZE: 256
+  NUM_WORKERS: 4
+
+INPUT:
+  SIZE: (224, 224)
+  INTERPOLATION: "bicubic"
+  PIXEL_MEAN: [0.48145466, 0.4578275, 0.40821073]
+  PIXEL_STD: [0.26862954, 0.26130258, 0.27577711]
+  TRANSFORMS: ["random_resized_crop", "random_flip", "normalize"]
+#  CUTOUT_N: 1
+#  CUTOUT_LEN: 128
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.0035
+  MAX_EPOCH: 5
+  LR_SCHEDULER: "cosine"
+  WARMUP_EPOCH: 1
+  WARMUP_TYPE: "constant"
+  WARMUP_CONS_LR: 1e-5
+
+TRAIN:
+  PRINT_FREQ: 20
+
+MODEL:
+  BACKBONE:
+    NAME: "ViT-B/16"
+
+
+
+TEST:
+  PER_CLASS_RESULT: false
+  FINAL_MODEL: "best_val"
+
+
+
+TRAINER:
+  MAPLE:
+    N_CTX: 2
+    CTX_INIT: "a photo of a"
+    PREC: "fp16"
+    PROMPT_DEPTH: 9

configs/trainers/MaPLe/vit_b16_c2_ep5_batch4_2ctx.yaml

@@ -0,0 +1,41 @@
+DATALOADER:
+  TRAIN_X:
+    BATCH_SIZE: 4
+  TEST:
+    BATCH_SIZE: 100
+  NUM_WORKERS: 4
+
+INPUT:
+  SIZE: (224, 224)
+  INTERPOLATION: "bicubic"
+  PIXEL_MEAN: [0.48145466, 0.4578275, 0.40821073]
+  PIXEL_STD: [0.26862954, 0.26130258, 0.27577711]
+  TRANSFORMS: ["random_resized_crop", "random_flip", "normalize"]
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.0035
+  MAX_EPOCH: 5
+  LR_SCHEDULER: "cosine"
+  WARMUP_EPOCH: 1
+  WARMUP_TYPE: "constant"
+  WARMUP_CONS_LR: 1e-5
+
+
+
+TRAIN:
+  PRINT_FREQ: 20
+
+TEST:
+  FINAL_MODEL: "best_val"
+
+MODEL:
+  BACKBONE:
+    NAME: "ViT-B/16"
+
+TRAINER:
+  MAPLE:
+    N_CTX: 2
+    CTX_INIT: "a photo of a"
+    PREC: "fp16"
+    PROMPT_DEPTH: 9

configs/trainers/MaPLe/vit_b16_c2_ep5_batch4_2ctx_cross_datasets.yaml

@@ -0,0 +1,36 @@
+DATALOADER:
+  TRAIN_X:
+    BATCH_SIZE: 4
+  TEST:
+    BATCH_SIZE: 100
+  NUM_WORKERS: 8
+
+INPUT:
+  SIZE: (224, 224)
+  INTERPOLATION: "bicubic"
+  PIXEL_MEAN: [0.48145466, 0.4578275, 0.40821073]
+  PIXEL_STD: [0.26862954, 0.26130258, 0.27577711]
+  TRANSFORMS: ["random_resized_crop", "random_flip", "normalize"]
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.0026
+  MAX_EPOCH: 2
+  LR_SCHEDULER: "cosine"
+  WARMUP_EPOCH: 1
+  WARMUP_TYPE: "constant"
+  WARMUP_CONS_LR: 1e-5
+
+TRAIN:
+  PRINT_FREQ: 20
+
+MODEL:
+  BACKBONE:
+    NAME: "ViT-B/16"
+
+TRAINER:
+  MAPLE:
+    N_CTX: 2
+    CTX_INIT: "a photo of a"
+    PREC: "fp16"
+    PROMPT_DEPTH: 9

configs/trainers/MaPLe/vit_b16_t.yaml

@@ -0,0 +1,53 @@
+DATASET:
+  SELECTION_BATCH_SIZE: 50
+  SUBSAMPLE_CLASSES: all
+
+
+DATALOADER:
+  RETURN_IMG0: true
+  TRAIN_X:
+    BATCH_SIZE: 4
+  TEST:
+    BATCH_SIZE: 128
+  NUM_WORKERS: 4
+
+INPUT:
+  SIZE: (224, 224)
+  INTERPOLATION: "bicubic"
+  PIXEL_MEAN: [0.48145466, 0.4578275, 0.40821073]
+  PIXEL_STD: [0.26862954, 0.26130258, 0.27577711]
+  TRANSFORMS: ["random_resized_crop", "random_flip", "normalize"]
+#  CUTOUT_N: 1
+#  CUTOUT_LEN: 128
+
+OPTIM:
+  NAME: "sgd"
+  LR: 0.0035 #0.0035     0.0026 for crossdata
+  MAX_EPOCH: 10
+  LR_SCHEDULER: "cosine"
+  WARMUP_EPOCH: 1
+  WARMUP_TYPE: "constant"
+  WARMUP_CONS_LR: 1e-5
+
+TRAIN:
+  PRINT_FREQ: 20
+  CHECKPOINT_FREQ: 1
+
+MODEL:
+  BACKBONE:
+    NAME: "ViT-B/16"
+
+
+
+TEST:
+  PER_CLASS_RESULT: false
+  FINAL_MODEL: "best_val"
+
+
+
+TRAINER:
+  MAPLE:
+    N_CTX: 2
+    CTX_INIT: "A photo of a"
+    PREC: "fp16"
+    PROMPT_DEPTH: 9