################################################################################ # Objects365 V2 · 365-class · YOLOv8-X Config (MMYOLO 0.6+) 2025-06-27 ################################################################################ # ────── A. 데이터 경로 ─────────────────────────────────────────────── DATA_ROOT = '/' TRAIN_JSON = '/data/ltb/ultralytics/objects365_mmyolo_coco/train_coco_fix.json' VAL_JSON = '/data/ltb/ultralytics/objects365_mmyolo_coco/val_coco_fix.json' LOAD_FROM = None # 사전 학습 weights 로딩 (없으면 None) load_from = LOAD_FROM # ────── B. 학습 관련 ─────────────────────────────────────────────── EPOCHS = 100 SWITCH_EPOCH = 80 CLOSE_MOSAIC_EPOCHS = 10 # ────── C. 배치 사이즈 ───────────────────────────────────────────── TRAIN_BATCH_SIZE = 8 VAL_BATCH_SIZE = 8 TEST_BATCH_SIZE = 8 # [D] 클래스 관련 CLASS_NAMES = ( # 1 – 20 "Person", "Sneakers", "Chair", "Other Shoes", "Hat", "Car", "Lamp", "Glasses", "Bottle", "Desk", "Cup", "Street Lights", "Cabinet/Shelf", "Handbag/Satchel", "Bracelet", "Plate", "Picture/Frame", "Helmet", "Book", "Gloves", # 21 – 40 "Storage box", "Boat", "Leather Shoes", "Flower", "Bench", "Potted Plant", "Bowl/Basin", "Flag", "Pillow", "Boots", "Vase", "Microphone", "Necklace", "Ring", "SUV", "Wine Glass", "Belt", "Moniter/TV", "Backpack", "Umbrella", # 41 – 60 "Traffic Light", "Speaker", "Watch", "Tie", "Trash bin Can", "Slippers", "Bicycle", "Stool", "Barrel/Bucket", "Van", "Couch", "Sandals", "Basket", "Drum", "Pen/Pencil", "Bus", "Wild Bird", "High Heels", "Motorcycle", "Guitar", # 61 – 80 "Carpet", "Cell Phone", "Bread", "Camera", "Canned", "Truck", "Traffic cone", "Cymbal", "Lifesaver", "Towel", "Stuffed Toy", "Candle", "Sailboat", "Laptop", "Awning", "Bed", "Faucet", "Tent", "Horse", "Mirror", # 81 – 100 "Power outlet", "Sink", "Apple", "Air Conditioner", "Knife", "Hockey Stick", "Paddle", "Pickup Truck", "Fork", "Traffic Sign", "Ballon", "Tripod", "Dog", "Spoon", "Clock", "Pot", "Cow", "Cake", "Dinning Table", "Sheep", # 101 – 120 "Hanger", "Blackboard/Whiteboard", "Napkin", "Other Fish", "Orange/Tangerine", "Toiletry", "Keyboard", "Tomato", "Lantern", "Machinery Vehicle", "Fan", "Green Vegetables", "Banana", "Baseball Glove", "Airplane", "Mouse", "Train", "Pumpkin", "Soccer", "Skiboard", # 121 – 140 "Luggage", "Nightstand", "Tea pot", "Telephone", "Trolley", "Head Phone", "Sports Car", "Stop Sign", "Dessert", "Scooter", "Stroller", "Crane", "Remote", "Refrigerator", "Oven", "Lemon", "Duck", "Baseball Bat", "Surveillance Camera", "Cat", # 141 – 160 "Jug", "Broccoli", "Piano", "Pizza", "Elephant", "Skateboard", "Surfboard", "Gun", "Skating and Skiing shoes", "Gas stove", "Donut", "Bow Tie", "Carrot", "Toilet", "Kite", "Strawberry", "Other Balls", "Shovel", "Pepper", "Computer Box", # 161 – 180 "Toilet Paper", "Cleaning Products", "Chopsticks", "Microwave", "Pigeon", "Baseball", "Cutting/Chopping Board", "Coffee Table", "Side Table", "Scissors", "Marker", "Pie", "Ladder", "Snowboard", "Cookies", "Radiator", "Fire Hydrant", "Basketball", "Zebra", "Grape", # 181 – 200 "Giraffe", "Potato", "Sausage", "Tricycle", "Violin", "Egg", "Fire Extinguisher", "Candy", "Fire Truck", "Billards", "Converter", "Bathtub", "Wheelchair", "Golf Club", "Briefcase", "Cucumber", "Cigar/Cigarette", "Paint Brush", "Pear", "Heavy Truck", # 201 – 220 "Hamburger", "Extractor", "Extention Cord", "Tong", "Tennis Racket", "Folder", "American Football", "Earphone", "Mask", "Kettle", "Tennis", "Ship", "Swing", "Coffee Machine", "Slide", "Carriage", "Onion", "Green beans", "Projector", "Frisbee", # 221 – 240 "Washing Machine/Drying Machine", "Chicken", "Printer", "Watermelon", "Saxophone", "Tissue", "Toothbrush", "Ice cream", "Hotair Ballon", "Cello", "French Fries", "Scale", "Trophy", "Cabbage", "Hot dog", "Blender", "Peach", "Rice", "Wallet/Purse", "Volleyball", # 241 – 260 "Deer", "Goose", "Tape", "Tablet", "Cosmetics", "Trumpet", "Pineapple", "Golf Ball", "Ambulance", "Parking meter", "Mango", "Key", "Hurdle", "Fishing Rod", "Medal", "Flute", "Brush", "Penguin", "Megaphone", "Corn", # 261 – 280 "Lettuce", "Garlic", "Swan", "Helicopter", "Green Onion", "Sandwich", "Nuts", "Speed Limit Sign", "Induction Cooker", "Broom", "Trombone", "Plum", "Rickshaw", "Goldfish", "Kiwi Fruit", "Router/Modem", "Poker Card", "Toaster", "Shrimp", "Sushi", # 281 – 300 "Cheese", "Notepaper", "Cherry", "Pliers", "CD", "Pasta", "Hammer", "Cue", "Avocado", "Hamimelon", "Flask", "Mushroom", "Screwdriver", "Soap", "Recorder", "Bear", "Eggplant", "Board Eraser", "Coconut", "Tape Measure/Ruler", # 301 – 320 "Pig", "Showerhead", "Globe", "Chips", "Steak", "Crosswalk Sign", "Stapler", "Camel", "Formula 1", "Pomegranate", "Dishwasher", "Crab", "Hoverboard", "Meat Ball", "Rice Cooker", "Tuba", "Calculator", "Papaya", "Antelope", "Parrot", # 321 – 340 "Seal", "Butterfly", "Dumbbell", "Donkey", "Lion", "Urinal", "Dolphin", "Electric Drill", "Hair Dryer", "Egg Tart", "Jellyfish", "Treadmill", "Lighter", "Grapefruit", "Game Board", "Mop", "Radish", "Baozi", "Target", "French", # 341 – 360 "Spring Rolls", "Monkey", "Rabbit", "Pencil Case", "Yak", "Red Cabbage", "Binoculars", "Asparagus", "Barbell", "Scallop", "Noodles", "Comb", "Dumpling", "Oyster", "Table Tennis Paddle", "Cosmetics Brush/Eyeliner Pencil", "Chainsaw", "Eraser", "Lobster", "Durian", # 361 – 365 "Okra", "Lipstick", "Cosmetics Mirror", "Curling", "Table Tennis" ) NUM_CLASSES = len(CLASS_NAMES) CLASS_WEIGHTS = [1.0] * NUM_CLASSES # ────── D. 하이퍼파라미터 ─────────────────────────────────────────── BASE_LR = 0.00125 MIXUP_PROB = 0.15 WEIGHT_DECAY = 0.0005 MOMENTUM = 0.937 ALBU_PROB = 0.01 PALETTE = [ (108, 60, 213), (127, 52, 97), (20, 41, 86), (180, 8, 173), (146, 110, 40), (225, 199, 203), (74, 47, 14), (212, 27, 157), (94, 154, 73), (230, 212, 62), (114, 131, 10), (45, 235, 253), (113, 65, 58), (118, 20, 133), (171, 134, 253), (14, 61, 58), (193, 185, 229), (27, 49, 35), (108, 236, 111), (92, 103, 75), (174, 145, 208), (232, 150, 1), (254, 58, 200), (12, 195, 158), (58, 156, 147), (94, 208, 44), (82, 80, 199), (152, 5, 187), (65, 29, 99), (132, 67, 171), (108, 22, 250), (40, 211, 158), (99, 128, 197), (63, 249, 37), (36, 185, 18), (249, 193, 187), (120, 94, 151), (222, 123, 74), (125, 139, 79), (137, 125, 134), (97, 28, 121), (84, 77, 13), (22, 136, 25), (5, 121, 254), (49, 88, 180), (233, 56, 96), (65, 144, 2), (156, 235, 35), (38, 80, 152), (130, 76, 81), (109, 9, 47), (129, 245, 11), (198, 206, 16), (188, 196, 158), (175, 200, 11), (222, 148, 248), (68, 81, 157), (176, 172, 27), (239, 107, 193), (125, 220, 10), (36, 158, 184), (169, 61, 137), (69, 231, 106), (82, 120, 0), (252, 209, 137), (87, 89, 118), (42, 38, 75), (234, 55, 174), (21, 58, 183), (53, 140, 104), (111, 100, 234), (21, 164, 102), (212, 54, 240), (184, 229, 4), (220, 214, 11), (203, 179, 11), (126, 120, 186), (250, 178, 89), (49, 153, 202), (199, 144, 53), (244, 246, 150), (170, 14, 82), (96, 96, 238), (145, 120, 219), (29, 133, 111), (201, 157, 71), (65, 47, 159), (69, 196, 166), (177, 239, 223), (27, 218, 72), (164, 142, 220), (193, 118, 237), (221, 143, 10), (231, 241, 229), (210, 119, 253), (48, 164, 70), (101, 223, 161), (217, 205, 133), (98, 42, 105), (237, 79, 163), (0, 180, 174), (136, 5, 67), (80, 197, 68), (182, 124, 138), (67, 130, 86), (24, 111, 142), (121, 193, 125), (176, 147, 162), (58, 199, 73), (231, 215, 96), (17, 247, 162), (106, 151, 134), (43, 87, 86), (167, 252, 221), (143, 49, 232), (102, 60, 158), (92, 112, 193), (242, 147, 204), (63, 59, 152), (218, 61, 159), (21, 204, 18), (101, 28, 34), (33, 20, 134), (37, 103, 131), (26, 221, 94), (21, 33, 154), (26, 76, 242), (136, 43, 5), (230, 141, 201), (226, 206, 38), (70, 30, 226), (224, 114, 251), (249, 62, 88), (53, 162, 176), (81, 246, 54), (117, 170, 84), (4, 139, 93), (95, 195, 159), (118, 89, 18), (112, 76, 184), (80, 140, 41), (82, 77, 176), (183, 185, 3), (164, 216, 9), (54, 236, 86), (116, 220, 108), (177, 226, 215), (189, 181, 108), (112, 143, 165), (7, 239, 102), (141, 192, 158), (146, 76, 161), (2, 3, 7), (189, 164, 106), (231, 59, 99), (18, 183, 194), (231, 94, 136), (111, 164, 24), (190, 66, 2), (189, 185, 6), (157, 16, 221), (216, 211, 96), (74, 183, 186), (15, 36, 162), (55, 252, 169), (113, 94, 38), (12, 55, 136), (170, 108, 141), (251, 27, 25), (246, 27, 181), (179, 109, 230), (180, 170, 222), (246, 12, 72), (44, 119, 64), (211, 47, 105), (45, 184, 78), (242, 189, 99), (93, 138, 15), (52, 134, 59), (112, 55, 34), (12, 24, 247), (147, 80, 38), (158, 229, 60), (187, 142, 137), (198, 136, 161), (128, 72, 171), (221, 152, 98), (20, 116, 173), (10, 81, 36), (228, 152, 159), (57, 85, 253), (52, 115, 159), (243, 229, 9), (147, 186, 7), (102, 183, 242), (252, 36, 137), (62, 65, 116), (142, 7, 194), (193, 131, 3), (4, 155, 56), (119, 136, 103), (187, 104, 26), (173, 40, 5), (108, 253, 187), (48, 91, 104), (249, 148, 106), (40, 74, 235), (27, 100, 30), (155, 0, 243), (69, 128, 141), (231, 85, 222), (124, 229, 212), (80, 187, 192), (29, 245, 95), (22, 27, 117), (170, 230, 186), (25, 57, 27), (22, 154, 7), (110, 195, 251), (154, 181, 143), (92, 51, 198), (8, 46, 56), (56, 160, 145), (156, 14, 7), (6, 143, 204), (191, 32, 59), (69, 58, 165), (145, 77, 115), (30, 19, 114), (133, 156, 219), (136, 143, 33), (184, 103, 243), (85, 105, 44), (69, 194, 203), (12, 221, 133), (208, 132, 154), (228, 163, 61), (65, 32, 112), (87, 70, 108), (71, 184, 164), (233, 71, 37), (29, 154, 166), (246, 69, 242), (18, 125, 191), (244, 1, 153), (137, 203, 207), (224, 8, 247), (33, 224, 78), (186, 24, 183), (29, 153, 222), (241, 221, 180), (189, 147, 211), (128, 71, 150), (33, 43, 147), (139, 216, 81), (112, 240, 163), (135, 220, 186), (19, 127, 3), (172, 105, 100), (179, 93, 110), (43, 198, 187), (140, 230, 30), (104, 221, 246), (195, 20, 61), (163, 113, 230), (243, 77, 43), (145, 53, 254), (86, 82, 220), (220, 161, 227), (177, 134, 101), (181, 123, 41), (213, 29, 215), (124, 42, 242), (53, 6, 206), (124, 245, 97), (186, 243, 247), (177, 119, 209), (114, 160, 55), (65, 232, 37), (134, 5, 74), (68, 151, 122), (93, 219, 81), (16, 151, 170), (17, 107, 195), (70, 83, 176), (88, 152, 215), (59, 181, 45), (120, 90, 193), (235, 247, 99), (253, 74, 168), (221, 44, 222), (96, 162, 95), (45, 221, 18), (213, 171, 104), (175, 173, 180), (119, 123, 169), (167, 131, 74), (50, 79, 99), (209, 151, 25), (20, 73, 108), (3, 250, 209), (145, 246, 36), (57, 230, 210), (229, 108, 123), (115, 11, 212), (142, 16, 76), (80, 33, 21), (26, 54, 81), (165, 189, 232), (111, 238, 1), (162, 137, 84), (108, 237, 209), (193, 11, 112), (10, 130, 8), (95, 114, 198), (40, 212, 195), (64, 24, 239), (105, 2, 4), (122, 150, 72), (37, 226, 68), (30, 118, 31), (225, 251, 113), (240, 82, 122), (64, 58, 125), (252, 151, 108), (212, 27, 233), (84, 113, 26), (5, 41, 248), (250, 5, 156), (65, 203, 79), (197, 125, 209), (248, 199, 43), (133, 202, 63), (155, 42, 6), (64, 163, 108), (161, 250, 44), (82, 2, 162), (231, 137, 206), (199, 188, 206), (132, 70, 180), (12, 137, 244), (215, 130, 76), (95, 144, 130), (90, 51, 25), (210, 9, 144), (73, 60, 251), (220, 233, 15), (15, 166, 219), (90, 172, 198), (76, 156, 252), (185, 202, 150), (158, 244, 177), (151, 136, 3), (253, 253, 252), (131, 152, 241), (97, 159, 190), (159, 38, 176), (114, 30, 79), (169, 142, 246), (7, 79, 105), (134, 141, 103), (68, 241, 10), (70, 247, 168), (195, 184, 125), (37, 230, 130), ] # ────── E. DataLoader 워커 수 ─────────────────────────────────────── NUM_WORKERS_TRAIN = 8 NUM_WORKERS_VAL = 8 NUM_WORKERS_TEST = 8 # ============================================================================= # 아래부터는 위 전역변수를 사용하여 전체 config를 구성합니다. # ============================================================================= _backend_args = None # ----- TTA (Test Time Augmentation) 위한 다양한 해상도 transform 설정 ----- _multiscale_resize_transforms = [ dict( type='Compose', transforms=[ dict(type='YOLOv5KeepRatioResize', scale=(640, 640)), dict( type='LetterResize', scale=(640, 640), allow_scale_up=False, pad_val=dict(img=114) ), ] ), dict( type='Compose', transforms=[ dict(type='YOLOv5KeepRatioResize', scale=(320, 320)), dict( type='LetterResize', scale=(320, 320), allow_scale_up=False, pad_val=dict(img=114) ), ] ), dict( type='Compose', transforms=[ dict(type='YOLOv5KeepRatioResize', scale=(960, 960)), dict( type='LetterResize', scale=(960, 960), allow_scale_up=False, pad_val=dict(img=114) ), ] ), ] affine_scale = 0.9 albu_train_transforms = [ dict(type='Blur', p=ALBU_PROB), dict(type='MedianBlur', p=ALBU_PROB), dict(type='ToGray', p=ALBU_PROB), dict(type='CLAHE', p=ALBU_PROB), ] backend_args = None base_lr = BASE_LR batch_shapes_cfg = None # ----- 클래스 이름과 팔레트 정의 (loss, 시각화 등에서 사용) ----- class_name = CLASS_NAMES close_mosaic_epochs = CLOSE_MOSAIC_EPOCHS # ----- custom_hooks: EMA, PipelineSwitchHook 등 ----- custom_hooks = [ dict( type='EMAHook', ema_type='ExpMomentumEMA', momentum=0.0001, update_buffers=True, priority=49, strict_load=False ), dict( type='mmdet.PipelineSwitchHook', switch_epoch=SWITCH_EPOCH, # 전역변수 사용 switch_pipeline=[ # 19epoch 이후에 사용할 파이프라인 예시 dict(type='LoadImageFromFile', backend_args=None), dict(type='LoadAnnotations', with_bbox=True), dict(type='YOLOv5KeepRatioResize', scale=(960, 960)), dict( type='LetterResize', scale=(960, 960), allow_scale_up=True, pad_val=dict(img=114.0) ), dict( type='YOLOv5RandomAffine', max_rotate_degree=10.0, max_shear_degree=2.0, scaling_ratio_range=(0.8, 1.2), border_val=(114, 114, 114), max_aspect_ratio=100 ), dict( type='mmdet.Albu', transforms=[ dict(type='Blur', p=ALBU_PROB), dict(type='MedianBlur', p=ALBU_PROB), dict(type='ToGray', p=ALBU_PROB), dict(type='CLAHE', p=ALBU_PROB), ], bbox_params=dict( format='pascal_voc', label_fields=['gt_bboxes_labels', 'gt_ignore_flags'], type='BboxParams' ), keymap=dict(img='image', gt_bboxes='bboxes'), ), dict(type='YOLOv5HSVRandomAug'), dict(type='mmdet.RandomFlip', prob=0.5), dict( type='mmdet.PackDetInputs', meta_keys=( 'img_id', 'img_path', 'ori_shape', 'img_shape', 'flip', 'flip_direction' ) ), ] ), ] data_root = DATA_ROOT # 전역변수 적용 dataset_type = 'YOLOv5CocoDataset' deepen_factor = 1.0 default_hooks = dict( checkpoint=dict( type='CheckpointHook', interval=1, max_keep_ckpts=10, save_best='auto' ), logger=dict(type='LoggerHook', interval=50), param_scheduler=dict( type='YOLOv5ParamSchedulerHook', scheduler_type='linear', max_epochs=EPOCHS, # 전역변수 사용 lr_factor=0.01 ), sampler_seed=dict(type='DistSamplerSeedHook'), timer=dict(type='IterTimerHook'), visualization=dict(type='mmdet.DetVisualizationHook') ) default_scope = 'mmyolo' env_cfg = dict( cudnn_benchmark=True, dist_cfg=dict(backend='nccl'), mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0) ) img_scale = (640, 640) img_scales = [(640, 640), (320, 320), (960, 960)] last_stage_out_channels = 512 last_transform = [ dict( type='mmdet.Albu', transforms=[ dict(type='Blur', p=ALBU_PROB), dict(type='MedianBlur', p=ALBU_PROB), dict(type='ToGray', p=ALBU_PROB), dict(type='CLAHE', p=ALBU_PROB), ], bbox_params=dict( format='pascal_voc', label_fields=['gt_bboxes_labels', 'gt_ignore_flags'], type='BboxParams' ), keymap=dict(img='image', gt_bboxes='bboxes'), ), dict(type='YOLOv5HSVRandomAug'), dict(type='mmdet.RandomFlip', prob=0.5), dict( type='mmdet.PackDetInputs', meta_keys=( 'img_id', 'img_path', 'ori_shape', 'img_shape', 'scale_factor', 'pad_param' ) ), ] launcher = 'none' load_config = dict( ignore_keys=[ 'bbox_head.head_module.cls_preds.0.weight', 'bbox_head.head_module.cls_preds.0.bias', 'bbox_head.head_module.cls_preds.1.weight', 'bbox_head.head_module.cls_preds.1.bias', 'bbox_head.head_module.cls_preds.2.weight', 'bbox_head.head_module.cls_preds.2.bias', ], strict=False ) load_from = LOAD_FROM log_level = 'INFO' log_processor = dict(type='LogProcessor', by_epoch=True, window_size=50) loss_bbox_weight = 7.5 loss_cls_weight = 1.0 loss_dfl_weight = 0.375 lr_factor = 0.01 max_aspect_ratio = 100 max_epochs = EPOCHS # 전역변수 사용 max_keep_ckpts = 2 metainfo = dict( classes=CLASS_NAMES, # 전역변수 적용 palette=PALETTE # 전역변수 적용 ) mixup_prob = MIXUP_PROB # ────── H. Model ──────────────────────────────────────────────────────── model = dict( type='YOLODetector', data_preprocessor=dict( type='YOLOv5DetDataPreprocessor', bgr_to_rgb=True, mean=[0.0, 0.0, 0.0], std=[255.0, 255.0, 255.0] ), backbone=dict( type='YOLOv8CSPDarknet', arch='P5', deepen_factor=deepen_factor, widen_factor=1.0, norm_cfg=dict(type='BN', momentum=0.03, eps=0.001), act_cfg=dict(type='SiLU', inplace=True), last_stage_out_channels=512 ), neck=dict( type='YOLOv8PAFPN', deepen_factor=deepen_factor, widen_factor=1.0, in_channels=[256, 512, 512], out_channels=[256, 512, 512], num_csp_blocks=3, norm_cfg=dict(type='BN', momentum=0.03, eps=0.001), act_cfg=dict(type='SiLU', inplace=True) ), bbox_head=dict( type='YOLOv8Head', bbox_coder=dict(type='DistancePointBBoxCoder'), head_module=dict( type='YOLOv8HeadModule', num_classes=NUM_CLASSES, # 전역변수 사용 in_channels=[256, 512, 512], featmap_strides=[8, 16, 32], reg_max=16, act_cfg=dict(type='SiLU', inplace=True), norm_cfg=dict(type='BN', momentum=0.03, eps=0.001) ), loss_bbox=dict( type='IoULoss', bbox_format='xywh', iou_mode='ciou', loss_weight=loss_bbox_weight, return_iou=False, reduction='sum' ), loss_cls=dict( type='CrossEntropyCustomLoss_mmyolo', use_sigmoid=True, num_classes=NUM_CLASSES, reduction='none', loss_weight=0.75, class_weight=CLASS_WEIGHTS # 전역변수 사용 ), loss_dfl=dict( type='mmdet.DistributionFocalLoss', reduction='mean', loss_weight=loss_dfl_weight ), prior_generator=dict( type='mmdet.MlvlPointGenerator', strides=[8, 16, 32], offset=0.5 ) ), train_cfg=dict( assigner=dict( type='BatchTaskAlignedAssigner', num_classes=NUM_CLASSES, topk=10, alpha=0.5, beta=6.0, eps=1e-09, use_ciou=True ) ), test_cfg=dict( nms=dict(type='nms', iou_threshold=0.5), nms_pre=30000, score_thr=0.01, max_per_img=300, multi_label=True ) ) model_test_cfg = dict( nms=dict(type='nms', iou_threshold=0.5), nms_pre=30000, score_thr=0.01, max_per_img=500, multi_label=True ) # ----- Mosaic transform 설정 예시 ----- mosaic_affine_transform = [ dict( type='Mosaic', img_scale=(640, 640), pad_val=114.0, pre_transform=[ dict(type='LoadImageFromFile', backend_args=None), dict(type='LoadAnnotations', with_bbox=True) ] ), dict( type='YOLOv5RandomAffine', border=(-320, -320), border_val=(114, 114, 114), max_aspect_ratio=100, max_rotate_degree=0.0, max_shear_degree=0.0, scaling_ratio_range=(0.1, 1.9) ), ] mosaic_affine_transform_960 = [ dict( type='Mosaic', img_scale=(960, 960), pad_val=114.0, pre_transform=[ dict(type='LoadImageFromFile', backend_args=None), dict(type='LoadAnnotations', with_bbox=True) ] ), dict( type='YOLOv5RandomAffine', border=(-480, -480), border_val=(114, 114, 114), max_aspect_ratio=100, max_rotate_degree=0.0, max_shear_degree=0.0, scaling_ratio_range=(0.1, 1.9) ), ] norm_cfg = dict(type='BN', momentum=0.03, eps=0.001) num_classes = NUM_CLASSES num_det_layers = 3 # ────── G. Optimizer & Scheduler ─────────────────────────────────────── optim_wrapper = dict( type='AmpOptimWrapper', constructor='YOLOv5OptimizerConstructor', loss_scale='dynamic', optimizer=dict( type='SGD', lr=BASE_LR, momentum=MOMENTUM, nesterov=True, weight_decay=WEIGHT_DECAY, batch_size_per_gpu=TRAIN_BATCH_SIZE ), clip_grad=dict(max_norm=5.0) ) param_scheduler = None pre_transform = [ dict(type='LoadImageFromFile', backend_args=None), dict(type='LoadAnnotations', with_bbox=True), ] resume = False save_epoch_intervals = 1 strides = [8, 16, 32] tal_alpha = 0.5 tal_beta = 6.0 tal_topk = 10 test_cfg = dict(type='TestLoop') test_dataloader = dict( batch_size=TEST_BATCH_SIZE, num_workers=NUM_WORKERS_TEST, persistent_workers=False, pin_memory=True, sampler=dict(type='DefaultSampler', shuffle=False), dataset=dict( type='YOLOv5CocoDataset', data_root=DATA_ROOT, ann_file=VAL_JSON, data_prefix=dict(img=''), metainfo=dict( classes=CLASS_NAMES, palette=PALETTE ), pipeline=[ dict(type='LoadImageFromFile'), dict(type='YOLOv5KeepRatioResize', scale=(960, 960)), dict( type='LetterResize', scale=(960, 960), allow_scale_up=False, pad_val=dict(img=114) ), # dict(type='mmdet.LoadAnnotations', with_bbox=True), dict( type='mmdet.PackDetInputs', meta_keys=( 'img_id', 'img_path', 'ori_shape', 'img_shape', 'scale_factor', 'pad_param' ) ), ], test_mode=True, batch_shapes_cfg=None ) ) test_evaluator = dict( type='mmdet.CocoMetric', ann_file= VAL_JSON, metric='bbox', proposal_nums=(100, 200, 500) ) test_pipeline = [ dict(type='LoadImageFromFile', backend_args=None), dict(type='YOLOv5KeepRatioResize', scale=(640, 640)), dict( type='LetterResize', scale=(640, 640), allow_scale_up=False, pad_val=dict(img=114) ), # dict(type='mmdet.LoadAnnotations', with_bbox=True), dict( type='mmdet.PackDetInputs', meta_keys=( 'img_id', 'img_path', 'ori_shape', 'img_shape', 'scale_factor', 'pad_param' ) ), ] train_ann_file = TRAIN_JSON train_batch_size_per_gpu = TRAIN_BATCH_SIZE train_cfg = dict( type='EpochBasedTrainLoop', max_epochs=EPOCHS, val_interval=1, dynamic_intervals=[(499, 1)] ) train_data_prefix = '' train_dataloader = dict( batch_size=TRAIN_BATCH_SIZE, num_workers=NUM_WORKERS_TRAIN, persistent_workers=False, pin_memory=True, sampler=dict(type='DefaultSampler', shuffle=True), collate_fn=dict(type='yolov5_collate'), dataset=dict( type='RepeatDataset', times=1, # 원본+회전캔버스 데이터 반복 횟수 (데이터 균형 조절) dataset=dict( type='YOLOv5CocoDataset', data_root=DATA_ROOT, ann_file=TRAIN_JSON, data_prefix=dict(img=''), filter_cfg=dict(filter_empty_gt=False, min_size=32), metainfo=dict( classes=CLASS_NAMES, palette=PALETTE ), pipeline=[ dict(type='LoadImageFromFile', backend_args=None), dict(type='LoadAnnotations', with_bbox=True), dict( type='YOLOv5RandomAffine', max_rotate_degree=10.0, max_shear_degree=2.0, scaling_ratio_range=(0.8, 1.2), border_val=(114, 114, 114) ), dict( type='Mosaic', img_scale=(960, 960), pad_val=114.0, pre_transform=[ dict(type='LoadImageFromFile', backend_args=None), dict(type='LoadAnnotations', with_bbox=True), ] ), dict( type='YOLOv5RandomAffine', border=(-480, -480), border_val=(114, 114, 114), max_aspect_ratio=100, max_rotate_degree=0.0, max_shear_degree=0.0, scaling_ratio_range=(0.1, 1.9) ), dict( type='mmdet.Albu', transforms=[ dict(type='Blur', p=ALBU_PROB), dict(type='MedianBlur', p=ALBU_PROB), dict(type='ToGray', p=ALBU_PROB), dict(type='CLAHE', p=ALBU_PROB), ], bbox_params=dict( type='BboxParams', format='pascal_voc', label_fields=['gt_bboxes_labels', 'gt_ignore_flags'] ), keymap=dict(img='image', gt_bboxes='bboxes') ), dict(type='YOLOv5HSVRandomAug'), dict(type='mmdet.RandomFlip', prob=0.5), dict( type='mmdet.PackDetInputs', meta_keys=( 'img_id', 'img_path', 'ori_shape', 'img_shape', 'flip', 'flip_direction' ) ), ], ) ), ) train_num_workers = NUM_WORKERS_TRAIN train_pipeline = [ # 학습 시 pipeline은 위와 동일하게 사용합니다. ] train_pipeline_stage2 = [ dict(type='LoadImageFromFile', backend_args=None), dict(type='LoadAnnotations', with_bbox=True), dict(type='YOLOv5KeepRatioResize', scale=(960, 960)), dict( type='LetterResize', scale=(960, 960), allow_scale_up=True, pad_val=dict(img=114.0) ), dict( type='YOLOv5RandomAffine', max_rotate_degree=10.0, max_shear_degree=2.0, scaling_ratio_range=(0.8, 1.2), border_val=(114, 114, 114), max_aspect_ratio=100 ), dict( type='mmdet.Albu', transforms=[ dict(type='Blur', p=ALBU_PROB), dict(type='MedianBlur', p=ALBU_PROB), dict(type='ToGray', p=ALBU_PROB), dict(type='CLAHE', p=ALBU_PROB), ], bbox_params=dict( type='BboxParams', format='pascal_voc', label_fields=['gt_bboxes_labels', 'gt_ignore_flags'] ), keymap=dict(img='image', gt_bboxes='bboxes') ), dict(type='YOLOv5HSVRandomAug'), dict(type='mmdet.RandomFlip', prob=0.5), dict( type='mmdet.PackDetInputs', meta_keys=( 'img_id', 'img_path', 'ori_shape', 'img_shape', 'flip', 'flip_direction' ) ), ] val_cfg = dict(type='ValLoop') val_data_prefix = '' val_dataloader = dict( batch_size=VAL_BATCH_SIZE, num_workers=NUM_WORKERS_VAL, persistent_workers=False, pin_memory=True, sampler=dict(type='DefaultSampler', shuffle=False), dataset=dict( type='YOLOv5CocoDataset', data_root=DATA_ROOT, ann_file=VAL_JSON, data_prefix=dict(img=''), metainfo=dict( classes=CLASS_NAMES, palette=PALETTE ), pipeline=[ dict(type='LoadImageFromFile'), dict(type='YOLOv5KeepRatioResize', scale=(960, 960)), dict( type='LetterResize', scale=(960, 960), allow_scale_up=False, pad_val=dict(img=114) ), # dict(type='mmdet.LoadAnnotations', with_bbox=True), dict( type='mmdet.PackDetInputs', meta_keys=( 'img_id', 'img_path', 'ori_shape', 'img_shape', 'scale_factor', 'pad_param' ) ), ], test_mode=True, batch_shapes_cfg=None ) ) val_evaluator = dict( type='mmdet.CocoMetric', ann_file= VAL_JSON, metric='bbox', proposal_nums=(100, 200, 500) ) val_interval_stage2 = 1 val_num_workers = NUM_WORKERS_VAL vis_backends = [dict(type='LocalVisBackend')] visualizer = dict( type='mmdet.DetLocalVisualizer', name='visualizer', vis_backends=[dict(type='LocalVisBackend')] ) weight_decay = WEIGHT_DECAY widen_factor = 1.0 work_dir = 'work_dirs/ff_base_cutmix_combined_1,2,5_01' tta_model = dict( type='mmdet.DetTTAModel', tta_cfg=dict( nms=dict(type='nms', iou_threshold=0.65), max_per_img=300 ) ) tta_pipeline = [ dict(type='LoadImageFromFile', backend_args=None), dict( type='TestTimeAug', transforms=[ [ dict( type='Compose', transforms=[ dict(type='YOLOv5KeepRatioResize', scale=(640, 640)), dict( type='LetterResize', scale=(640, 640), allow_scale_up=False, pad_val=dict(img=114) ) ] ), dict( type='Compose', transforms=[ dict(type='YOLOv5KeepRatioResize', scale=(320, 320)), dict( type='LetterResize', scale=(320, 320), allow_scale_up=False, pad_val=dict(img=114) ) ] ), dict( type='Compose', transforms=[ dict(type='YOLOv5KeepRatioResize', scale=(960, 960)), dict( type='LetterResize', scale=(960, 960), allow_scale_up=False, pad_val=dict(img=114) ) ] ), ], [ dict(type='mmdet.RandomFlip', prob=1.0), dict(type='mmdet.RandomFlip', prob=0.0), ], [ dict(type='mmdet.LoadAnnotations', with_bbox=True), ], [ dict( type='mmdet.PackDetInputs', meta_keys=( 'img_id', 'img_path', 'ori_shape', 'img_shape', 'scale_factor', 'pad_param', 'flip', 'flip_direction' ) ), ], ] ), ]