First commit.

configs/R_50/TotalText/finetune_150k_tt_mlt_13_15_textocr.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "output/R50/150k_tt_mlt_13_15_textocr/pretrain/model_final.pth"
+
+DATASETS:
+  TRAIN: ("totaltext_train",)
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-5
+  LR_BACKBONE: 1e-6
+  WARMUP_ITERS: 0
+  STEPS: (100000,)  # no step
+  MAX_ITER: 2000
+  CHECKPOINT_PERIOD: 2000
+
+TEST:
+  EVAL_PERIOD: 1000
+
+OUTPUT_DIR: "output/R50/150k_tt_mlt_13_15_textocr/finetune/totaltext"
\ No newline at end of file

configs/R_50/pretrain/150k_tt.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "detectron2://ImageNetPretrained/torchvision/R-50.pkl"
+  
+DATASETS:
+  TRAIN: ("syntext1","syntext2","totaltext_train",)
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-4
+  LR_BACKBONE: 1e-5
+  WARMUP_ITERS: 0
+  STEPS: (300000,)
+  MAX_ITER: 350000
+  CHECKPOINT_PERIOD: 100000
+
+TEST:
+  EVAL_PERIOD: 10000
+
+OUTPUT_DIR: "output/R50/150k_tt/pretrain"
\ No newline at end of file

configs/R_50/pretrain/150k_tt_mlt_13_15.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "detectron2://ImageNetPretrained/torchvision/R-50.pkl"
+  
+DATASETS:
+  TRAIN: ("syntext1","syntext2","totaltext_train","mlt","ic13_train","ic15_train",)
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-4
+  LR_BACKBONE: 1e-5
+  WARMUP_ITERS: 0
+  STEPS: (320000,)
+  MAX_ITER: 375000
+  CHECKPOINT_PERIOD: 100000
+
+TEST:
+  EVAL_PERIOD: 10000
+
+OUTPUT_DIR: "output/R50/150k_tt_mlt_13_15/pretrain"
\ No newline at end of file

configs/R_50/pretrain/150k_tt_mlt_13_15_textocr.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "detectron2://ImageNetPretrained/torchvision/R-50.pkl"
+
+DATASETS:
+  TRAIN: ("syntext1","syntext2","totaltext_train","mlt","ic13_train","ic15_train","textocr1","textocr2",)
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-4
+  LR_BACKBONE: 1e-5
+  WARMUP_ITERS: 0
+  STEPS: (375000,)
+  MAX_ITER: 435000
+  CHECKPOINT_PERIOD: 100000
+
+TEST:
+  EVAL_PERIOD: 10000
+
+OUTPUT_DIR: "output/R50/150k_tt_mlt_13_15_textocr/pretrain"
\ No newline at end of file

configs/Swin_S/Base_det.yaml

+MODEL:
+  META_ARCHITECTURE: "TransformerPureDetector"
+  MASK_ON: False
+  PIXEL_MEAN: [123.675, 116.280, 103.530]
+  PIXEL_STD: [58.395, 57.120, 57.375]
+  BACKBONE:
+    NAME: "build_swin_backbone"
+    FREEZE_AT: -1
+  SWIN:
+    TYPE: 'small'
+    DROP_PATH_RATE: 0.3
+  TRANSFORMER:
+    ENABLED: True
+    NUM_FEATURE_LEVELS: 4
+    TEMPERATURE: 10000
+    ENC_LAYERS: 6
+    DEC_LAYERS: 6
+    DIM_FEEDFORWARD: 1024
+    HIDDEN_DIM: 256
+    DROPOUT: 0.0
+    NHEADS: 8
+    NUM_QUERIES: 100
+    ENC_N_POINTS: 4
+    DEC_N_POINTS: 4
+    NUM_POINTS: 25
+    INFERENCE_TH_TEST: 0.4
+    LOSS:
+      BEZIER_SAMPLE_POINTS: 25
+      BEZIER_CLASS_WEIGHT: 1.0
+      BEZIER_COORD_WEIGHT: 1.0
+      POINT_CLASS_WEIGHT: 1.0
+      POINT_COORD_WEIGHT: 1.0
+      POINT_TEXT_WEIGHT: 0.5
+      BOUNDARY_WEIGHT: 0.5
+
+SOLVER:
+  WEIGHT_DECAY: 1e-4
+  OPTIMIZER: "ADAMW"
+  LR_BACKBONE_NAMES: ['backbone.0']
+  LR_LINEAR_PROJ_NAMES: ['reference_points', 'sampling_offsets']
+  LR_LINEAR_PROJ_MULT: 1.
+  CLIP_GRADIENTS:
+    ENABLED: True
+    CLIP_TYPE: "full_model"
+    CLIP_VALUE: 0.1
+    NORM_TYPE: 2.0
+
+INPUT:
+  HFLIP_TRAIN: False
+  MIN_SIZE_TRAIN: (480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800, 832, 864, 896)
+  MAX_SIZE_TRAIN: 1600
+  MIN_SIZE_TEST: 1000
+  MAX_SIZE_TEST: 1892
+  CROP:
+    ENABLED: True
+    CROP_INSTANCE: False
+    SIZE: [0.1, 0.1]
+  FORMAT: "RGB"
+
+DATALOADER:
+  NUM_WORKERS: 8
+
+VERSION: 2
+SEED: 42
\ No newline at end of file

configs/Swin_S/TotalText/finetune_150k_tt_mlt_13_15.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "output/SwinS/150k_tt_mlt_13_15/pretrain/model_final.pth"
+
+DATASETS:
+  TRAIN: ("totaltext_train",)
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-5
+  LR_BACKBONE: 1e-5
+  WARMUP_ITERS: 0
+  STEPS: (100000,)  # no step
+  MAX_ITER: 10000
+  CHECKPOINT_PERIOD: 10000
+
+TEST:
+  EVAL_PERIOD: 1000
+
+OUTPUT_DIR: "output/SwinS/150k_tt_mlt_13_15/finetune/totaltext"
\ No newline at end of file

configs/Swin_S/pretrain/150k_tt_mlt_13_15.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "./pretrained_backbone/swin_small_patch4_window7_224_convert.pth"
+
+DATASETS:
+  TRAIN: ("syntext1","syntext2","totaltext_train","mlt","ic13_train","ic15_train",)
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-4
+  LR_BACKBONE: 1e-4
+  WARMUP_ITERS: 0
+  STEPS: (320000,)
+  MAX_ITER: 375000
+  CHECKPOINT_PERIOD: 100000
+
+TEST:
+  EVAL_PERIOD: 10000
+
+OUTPUT_DIR: "output/SwinS/150k_tt_mlt_13_15/pretrain"
\ No newline at end of file

configs/Swin_T/Base_det.yaml

+MODEL:
+  META_ARCHITECTURE: "TransformerPureDetector"
+  MASK_ON: False
+  PIXEL_MEAN: [123.675, 116.280, 103.530]
+  PIXEL_STD: [58.395, 57.120, 57.375]
+  BACKBONE:
+    NAME: "build_swin_backbone"
+    FREEZE_AT: -1
+  SWIN:
+    TYPE: 'tiny'
+    DROP_PATH_RATE: 0.2
+  TRANSFORMER:
+    ENABLED: True
+    NUM_FEATURE_LEVELS: 4
+    TEMPERATURE: 10000
+    ENC_LAYERS: 6
+    DEC_LAYERS: 6
+    DIM_FEEDFORWARD: 1024
+    HIDDEN_DIM: 256
+    DROPOUT: 0.0
+    NHEADS: 8
+    NUM_QUERIES: 100
+    ENC_N_POINTS: 4
+    DEC_N_POINTS: 4
+    NUM_POINTS: 25
+    INFERENCE_TH_TEST: 0.4
+    LOSS:
+      BEZIER_SAMPLE_POINTS: 25
+      BEZIER_CLASS_WEIGHT: 1.0
+      BEZIER_COORD_WEIGHT: 1.0
+      POINT_CLASS_WEIGHT: 1.0
+      POINT_COORD_WEIGHT: 1.0
+      POINT_TEXT_WEIGHT: 0.5
+      BOUNDARY_WEIGHT: 0.5
+
+SOLVER:
+  WEIGHT_DECAY: 1e-4
+  OPTIMIZER: "ADAMW"
+  LR_BACKBONE_NAMES: ['backbone.0']
+  LR_LINEAR_PROJ_NAMES: ['reference_points', 'sampling_offsets']
+  LR_LINEAR_PROJ_MULT: 1.
+  CLIP_GRADIENTS:
+    ENABLED: True
+    CLIP_TYPE: "full_model"
+    CLIP_VALUE: 0.1
+    NORM_TYPE: 2.0
+
+INPUT:
+  HFLIP_TRAIN: False
+  MIN_SIZE_TRAIN: (480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800, 832, 864, 896)
+  MAX_SIZE_TRAIN: 1600
+  MIN_SIZE_TEST: 1000
+  MAX_SIZE_TEST: 1892
+  CROP:
+    ENABLED: True
+    CROP_INSTANCE: False
+    SIZE: [0.1, 0.1]
+  FORMAT: "RGB"
+
+DATALOADER:
+  NUM_WORKERS: 8
+
+VERSION: 2
+SEED: 42
\ No newline at end of file

configs/Swin_T/TotalText/finetune_150k_tt_mlt_13_15.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "output/SwinT/150k_tt_mlt_13_15/pretrain/model_final.pth"
+
+DATASETS:
+  TRAIN: ("totaltext_train",)
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-5
+  LR_BACKBONE: 1e-5
+  WARMUP_ITERS: 0
+  STEPS: (100000,)  # no step
+  MAX_ITER: 10000
+  CHECKPOINT_PERIOD: 2000
+
+TEST:
+  EVAL_PERIOD: 1000
+
+OUTPUT_DIR: "output/SwinT/150k_tt_mlt_13_15/finetune/totaltext"
\ No newline at end of file

configs/Swin_T/pretrain/150k_tt_mlt_13_15.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "./pretrained_backbone/swin_tiny_patch4_window7_224_convert.pth"
+
+DATASETS:
+  TRAIN: ("syntext1","syntext2","totaltext_train","mlt","ic13_train","ic15_train",)
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-4
+  LR_BACKBONE: 1e-4
+  WARMUP_ITERS: 0
+  STEPS: (320000,)
+  MAX_ITER: 375000
+  CHECKPOINT_PERIOD: 100000
+
+TEST:
+  EVAL_PERIOD: 10000
+
+OUTPUT_DIR: "output/SwinT/150k_tt_mlt_13_15/pretrain"
\ No newline at end of file

configs/ViTAEv2_S/Base_det.yaml

+MODEL:
+  META_ARCHITECTURE: "TransformerPureDetector"
+  MASK_ON: False
+  PIXEL_MEAN: [123.675, 116.280, 103.530]
+  PIXEL_STD: [58.395, 57.120, 57.375]
+  BACKBONE:
+    NAME: "build_vitaev2_backbone"
+  ViTAEv2:
+    TYPE: 'vitaev2_s'
+    DROP_PATH_RATE: 0.3
+  TRANSFORMER:
+    ENABLED: True
+    NUM_FEATURE_LEVELS: 4
+    TEMPERATURE: 10000
+    ENC_LAYERS: 6
+    DEC_LAYERS: 6
+    DIM_FEEDFORWARD: 1024
+    HIDDEN_DIM: 256
+    DROPOUT: 0.0
+    NHEADS: 8
+    NUM_QUERIES: 100
+    ENC_N_POINTS: 4
+    DEC_N_POINTS: 4
+    NUM_POINTS: 25
+    INFERENCE_TH_TEST: 0.4
+    LOSS:
+      BEZIER_SAMPLE_POINTS: 25
+      BEZIER_CLASS_WEIGHT: 1.0
+      BEZIER_COORD_WEIGHT: 1.0
+      POINT_CLASS_WEIGHT: 1.0
+      POINT_COORD_WEIGHT: 1.0
+      POINT_TEXT_WEIGHT: 0.5
+      BOUNDARY_WEIGHT: 0.5
+
+SOLVER:
+  WEIGHT_DECAY: 1e-4
+  OPTIMIZER: "ADAMW"
+  LR_BACKBONE_NAMES: ['backbone.0']
+  LR_LINEAR_PROJ_NAMES: ['reference_points', 'sampling_offsets']
+  LR_LINEAR_PROJ_MULT: 1.
+  CLIP_GRADIENTS:
+    ENABLED: True
+    CLIP_TYPE: "full_model"
+    CLIP_VALUE: 0.1
+    NORM_TYPE: 2.0
+
+INPUT:
+  HFLIP_TRAIN: False
+  MIN_SIZE_TRAIN: (480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800, 832, 864, 896)
+  MAX_SIZE_TRAIN: 1600
+  MIN_SIZE_TEST: 1024
+  MAX_SIZE_TEST: 1892
+  CROP:
+    ENABLED: True
+    CROP_INSTANCE: False
+    SIZE: [0.1, 0.1]
+  FORMAT: "RGB"
+
+DATALOADER:
+  NUM_WORKERS: 8
+
+VERSION: 2
+SEED: 42
\ No newline at end of file

configs/ViTAEv2_S/IC15/finetune_150k_tt_mlt_13_15.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "output/vitaev2_s/150k_tt_mlt_13_15/pretrain/model_final.pth"
+  ViTAEv2:
+    DROP_PATH_RATE: 0.2
+  TRANSFORMER:
+    INFERENCE_TH_TEST: 0.3
+
+DATASETS:
+  TRAIN: ("ic15_train",)
+  TEST: ("ic15_test",)
+
+INPUT:
+  MIN_SIZE_TRAIN: (800,900,1000,1100,1200,1300,1400)
+  MAX_SIZE_TRAIN: 3000
+  MIN_SIZE_TEST: 1440
+  MAX_SIZE_TEST: 4000
+  CROP:
+    ENABLED: False
+  ROTATE: False
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-5
+  LR_BACKBONE: 1e-5
+  WARMUP_ITERS: 0
+  STEPS: (100000,)
+  MAX_ITER: 3000
+  CHECKPOINT_PERIOD: 3000
+
+TEST:
+  EVAL_PERIOD: 1000
+  # 1 - Generic, 2 - Weak, 3 - Strong (for icdar2015)
+  LEXICON_TYPE: 3
+
+OUTPUT_DIR: "output/vitaev2_s/150k_tt_mlt_13_15/finetune/ic15"
\ No newline at end of file

configs/ViTAEv2_S/IC15/finetune_150k_tt_mlt_13_15_textocr.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "output/vitaev2_s/150k_tt_mlt_13_15_textocr/pretrain/model_final.pth"
+  ViTAEv2:
+    DROP_PATH_RATE: 0.2
+  TRANSFORMER:
+    INFERENCE_TH_TEST: 0.3
+
+DATASETS:
+  TRAIN: ("ic15_train",)
+  TEST: ("ic15_test",)
+
+INPUT:
+  MIN_SIZE_TRAIN: (800,900,1000,1100,1200,1300,1400)
+  MAX_SIZE_TRAIN: 3000
+  MIN_SIZE_TEST: 1440
+  MAX_SIZE_TEST: 4000
+  CROP:
+    ENABLED: False
+  ROTATE: False
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-5
+  LR_BACKBONE: 1e-5
+  WARMUP_ITERS: 0
+  STEPS: (100000,)
+  MAX_ITER: 1000
+  CHECKPOINT_PERIOD: 1000
+
+TEST:
+  EVAL_PERIOD: 500
+  # 1 - Generic, 2 - Weak, 3 - Strong (for icdar2015)
+  LEXICON_TYPE: 3
+
+OUTPUT_DIR: "output/vitaev2_s/150k_tt_mlt_13_15_textocr/finetune/ic15"
\ No newline at end of file

configs/ViTAEv2_S/ReCTS/finetune.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "./output/vitaev2_s/rects/pretrain/model_final.pth"
+  ViTAEv2:
+    DROP_PATH_RATE: 0.2
+  TRANSFORMER:
+    VOC_SIZE: 5462
+    CUSTOM_DICT: "chn_cls_list"
+    INFERENCE_TH_TEST: 0.35
+    LOSS:
+      POINT_TEXT_WEIGHT: 1.0
+  
+DATASETS:
+  TRAIN: ("rects_train", "rects_val",)
+  TEST: ("rects_test",)
+INPUT:
+  ROTATE: False
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-5
+  LR_BACKBONE: 1e-5
+  WARMUP_ITERS: 0
+  STEPS: (20000,)
+  MAX_ITER: 30000
+  CHECKPOINT_PERIOD: 30000
+
+TEST:
+  EVAL_PERIOD: 100000000
+
+OUTPUT_DIR: "output/vitaev2_s/rects/finetune"
\ No newline at end of file

configs/ViTAEv2_S/ReCTS/pretrain.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "./pretrained_backbone/vitaev2_s_convert.pth"
+  TRANSFORMER:
+    VOC_SIZE: 5462
+  
+DATASETS:
+  TRAIN: ("chnsyn_train", "rects_train", "rects_val", "lsvt_train", "art_train",)
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-4
+  LR_BACKBONE: 1e-4
+  WARMUP_ITERS: 0
+  STEPS: (300000,)
+  MAX_ITER: 400000
+  CHECKPOINT_PERIOD: 300000
+
+TEST:
+  EVAL_PERIOD: 100000000
+
+OUTPUT_DIR: "output/vitaev2_s/rects/pretrain"

configs/ViTAEv2_S/TotalText/finetune_150k_tt_mlt_13_15.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "output/vitaev2_s/150k_tt_mlt_13_15/pretrain/model_final.pth"
+  ViTAEv2:
+    DROP_PATH_RATE: 0.2
+
+DATASETS:
+  TRAIN: ("totaltext_train",)
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-5
+  LR_BACKBONE: 1e-5
+  WARMUP_ITERS: 0
+  STEPS: (100000,)
+  MAX_ITER: 10000
+  CHECKPOINT_PERIOD: 10000
+
+TEST:
+  EVAL_PERIOD: 1000
+
+OUTPUT_DIR: "output/vitaev2_s/150k_tt_mlt_13_15/finetune/totaltext"
\ No newline at end of file

configs/ViTAEv2_S/TotalText/finetune_150k_tt_mlt_13_15_textocr.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "output/vitaev2_s/150k_tt_mlt_13_15_textocr/pretrain/model_final.pth"
+  ViTAEv2:
+    DROP_PATH_RATE: 0.2
+
+DATASETS:
+  TRAIN: ("totaltext_train",)
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-5
+  LR_BACKBONE: 1e-5
+  WARMUP_ITERS: 0
+  STEPS: (100000,)
+  MAX_ITER: 2000
+  CHECKPOINT_PERIOD: 2000
+
+TEST:
+  EVAL_PERIOD: 1000
+
+OUTPUT_DIR: "output/vitaev2_s/150k_tt_mlt_13_15_textocr/finetune/totaltext"
\ No newline at end of file

configs/ViTAEv2_S/pretrain/150k_tt_mlt_13_15.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "pretrained_backbone/vitaev2_s_convert.pth"
+
+DATASETS:
+  TRAIN: ("syntext1","syntext2","totaltext_train","mlt","ic13_train","ic15_train",)
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-4
+  LR_BACKBONE: 1e-4
+  WARMUP_ITERS: 0
+  STEPS: (320000,)
+  MAX_ITER: 375000
+  CHECKPOINT_PERIOD: 100000
+
+TEST:
+  EVAL_PERIOD: 10000
+
+OUTPUT_DIR: "output/vitaev2_s/150k_tt_mlt_13_15/pretrain"
\ No newline at end of file

configs/ViTAEv2_S/pretrain/150k_tt_mlt_13_15_textocr.yaml

+_BASE_: "../Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "pretrained_backbone/vitaev2_s_convert.pth"
+
+DATASETS:
+  TRAIN: ("syntext1","syntext2","totaltext_train","mlt","ic13_train","ic15_train","textocr1","textocr2")
+  TEST: ("totaltext_test",)
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 1e-4
+  LR_BACKBONE: 1e-4
+  WARMUP_ITERS: 0
+  STEPS: (375000,)
+  MAX_ITER: 435000
+  CHECKPOINT_PERIOD: 100000
+
+TEST:
+  EVAL_PERIOD: 10000
+
+OUTPUT_DIR: "output/vitaev2_s/150k_tt_mlt_13_15_textocr/pretrain"
\ No newline at end of file

configs/simple/test_simple.yaml

+_BASE_: "../R_50/Base_det.yaml"
+
+MODEL:
+  WEIGHTS: "pretrained_models/CTW1500/finetune_ctw_96voc.pth" # 可替换为自己的预训练模型地址
+  TRANSFORMER:
+    VOC_SIZE: 96
+    NUM_POINTS: 50
+    LOSS:
+        BEZIER_SAMPLE_POINTS: 50  # the same as NUM_POINTS
+        BEZIER_CLASS_WEIGHT: 1.0
+        BEZIER_COORD_WEIGHT: 0.5
+        POINT_CLASS_WEIGHT: 1.0
+        POINT_COORD_WEIGHT: 0.5
+        POINT_TEXT_WEIGHT: 1.0
+        BOUNDARY_WEIGHT: 0.25
+
+DATASETS:
+  TRAIN: ("simple_train",)
+  TEST: ("simple_test",)
+
+INPUT:
+  MIN_SIZE_TEST: 1000
+  MAX_SIZE_TEST: 1200
+
+SOLVER:
+  IMS_PER_BATCH: 8
+  BASE_LR: 5e-5
+  LR_BACKBONE: 5e-6
+  WARMUP_ITERS: 0
+  STEPS: (8000,)
+  MAX_ITER: 12000
+  CHECKPOINT_PERIOD: 4000
+
+TEST:
+  EVAL_PERIOD: 1000
+
+OUTPUT_DIR: "output/R50/simple/pretrain_maxlen50_96voc"