feat: 초기 프로젝트 구조 추가

ai/__init__.py

@@ -0,0 +1,3 @@
+# ai 패키지 — Detector, Trainer 노출
+from .detector import Detector
+from .trainer import Trainer

ai/dataset/data.yaml

@@ -0,0 +1,9 @@
+names:
+- 스크래치
+- 이물
+- 흑점
+- 변형
+nc: 4
+path: e:\ANT\ai\dataset
+train: images/train
+val: images/val

ai/dataset/labels/train/스크린샷 2026-04-27 175539.txt

@@ -0,0 +1 @@
+0 0.233974 0.409524 0.339744 0.590476

ai/dataset/labels/train/스크린샷 2026-05-07 100837.txt

@@ -0,0 +1 @@
+0 0.318598 0.542781 0.472561 0.754011

ai/dataset/labels/val/스크린샷 2026-04-27 175231.txt

@@ -0,0 +1,2 @@
+0 0.227074 0.195219 0.401747 0.215139
+1 0.853712 0.252988 0.257642 0.330677

ai/detector.py

@@ -0,0 +1,59 @@
+# AI 추론 — YOLOv8 기반 불량(스크래치/이물/흑점/변형) 검출
+import os
+import numpy as np
+
+from utils.path_helper import BASE_PATH
+
+
+class Detector:
+    class_names = ["스크래치", "이물", "흑점", "변형"]
+
+    def __init__(self):
+        self._model     = None
+        self.model_path = None
+
+    def load_model(self, model_path: str) -> bool:
+        if model_path and not os.path.isabs(model_path):
+            model_path = os.path.join(BASE_PATH, model_path)
+        try:
+            from ultralytics import YOLO  # 지연 로딩 — 앱 시작 시 torch DLL 오류 방지
+            self._model     = YOLO(model_path)
+            self.model_path = model_path
+            print(f"[AI] 모델 로드 성공: {model_path}")
+            return True
+        except Exception as e:
+            print(f"[AI] 모델 로드 실패: {e}")
+            self._model = None
+            return False
+
+    def is_loaded(self) -> bool:
+        return self._model is not None
+
+    def detect(self, image: np.ndarray) -> list:
+        """
+        image: numpy array (BGR)
+        반환: [{"class_id": int, "class_name": str,
+                "confidence": float, "bbox": [x1,y1,x2,y2]}, ...]
+        """
+        if self._model is None:
+            return []
+        try:
+            results = self._model(image, verbose=False)
+            detections = []
+            for result in results:
+                for box in result.boxes:
+                    class_id = int(box.cls[0])
+                    detections.append({
+                        "class_id":   class_id,
+                        "class_name": (
+                            self.class_names[class_id]
+                            if class_id < len(self.class_names)
+                            else str(class_id)
+                        ),
+                        "confidence": float(box.conf[0]),
+                        "bbox":       box.xyxy[0].tolist(),
+                    })
+            return detections
+        except Exception as e:
+            print(f"[AI] 추론 오류: {e}")
+            return []

ai/runs/train/args.yaml

@@ -0,0 +1,110 @@
+task: detect
+mode: train
+model: yolov8n.pt
+data: e:\ANT\ai\dataset\data.yaml
+epochs: 100
+time: null
+patience: 100
+batch: 16
+imgsz: 640
+save: true
+save_period: -1
+cache: false
+device: null
+workers: 0
+project: e:\ANT\ai\runs
+name: train
+exist_ok: true
+pretrained: true
+optimizer: auto
+verbose: true
+seed: 0
+deterministic: true
+single_cls: false
+rect: false
+cos_lr: false
+close_mosaic: 10
+resume: false
+amp: false
+fraction: 1.0
+profile: false
+freeze: null
+multi_scale: 0.0
+compile: false
+overlap_mask: true
+mask_ratio: 4
+dropout: 0.0
+val: true
+split: val
+save_json: false
+conf: null
+iou: 0.7
+max_det: 300
+half: false
+dnn: false
+plots: false
+end2end: null
+source: null
+vid_stride: 1
+stream_buffer: false
+visualize: false
+augment: false
+agnostic_nms: false
+classes: null
+retina_masks: false
+embed: null
+show: false
+save_frames: false
+save_txt: false
+save_conf: false
+save_crop: false
+show_labels: true
+show_conf: true
+show_boxes: true
+line_width: null
+format: torchscript
+keras: false
+optimize: false
+int8: false
+dynamic: false
+simplify: true
+opset: null
+workspace: null
+nms: false
+lr0: 0.01
+lrf: 0.01
+momentum: 0.937
+weight_decay: 0.0005
+warmup_epochs: 3.0
+warmup_momentum: 0.8
+warmup_bias_lr: 0.1
+box: 7.5
+cls: 0.5
+cls_pw: 0.0
+dfl: 1.5
+pose: 12.0
+kobj: 1.0
+rle: 1.0
+angle: 1.0
+nbs: 64
+hsv_h: 0.015
+hsv_s: 0.7
+hsv_v: 0.4
+degrees: 0.0
+translate: 0.1
+scale: 0.5
+shear: 0.0
+perspective: 0.0
+flipud: 0.0
+fliplr: 0.5
+bgr: 0.0
+mosaic: 1.0
+mixup: 0.0
+cutmix: 0.0
+copy_paste: 0.0
+copy_paste_mode: flip
+auto_augment: randaugment
+erasing: 0.4
+cfg: null
+tracker: botsort.yaml
+save_dir: E:\ANT\ai\runs\train

ai/runs/train/results.csv

@@ -0,0 +1,101 @@
+epoch,time,train/box_loss,train/cls_loss,train/dfl_loss,metrics/precision(B),metrics/recall(B),metrics/mAP50(B),metrics/mAP50-95(B),val/box_loss,val/cls_loss,val/dfl_loss,lr/pg0,lr/pg1,lr/pg2
+1,0.744601,2.4784,4.2576,2.37292,0.00581,0.5,0.00599,0.00178,2.08106,4.5092,2.3941,0,0,0
+2,1.17594,3.54636,4.89413,2.76008,0.00588,0.5,0.00622,0.00124,2.10159,4.51392,2.40401,1.23763e-05,1.23763e-05,1.23763e-05
+3,1.4434,3.2212,4.47996,3.91873,0.01514,1,0.01568,0.0022,2.24221,4.37976,2.4629,2.4505e-05,2.4505e-05,2.4505e-05
+4,1.81097,3.23832,4.18544,3.38558,0.01459,1,0.01534,0.0037,2.32948,4.47892,2.46147,3.63862e-05,3.63862e-05,3.63862e-05
+5,2.17724,2.76973,4.43589,2.74067,0.00556,0.5,0.00711,0.00195,2.33679,4.46132,2.45867,4.802e-05,4.802e-05,4.802e-05
+6,2.47758,2.93916,4.25681,2.81574,0.00562,0.5,0.00711,0.00197,2.23356,4.4439,2.55513,5.94062e-05,5.94062e-05,5.94062e-05
+7,2.77856,2.95798,3.80506,2.41629,0.01433,1,0.0167,0.00167,2.13047,4.47684,2.46296,7.0545e-05,7.0545e-05,7.0545e-05
+8,3.08538,2.97272,3.8706,2.66583,0.00847,0.5,0.00939,0.00094,1.93983,4.4964,2.44096,8.14362e-05,8.14362e-05,8.14362e-05
+9,3.38628,1.96121,4.27233,2.04591,0,0,0,0,1.67434,4.64554,2.3051,9.208e-05,9.208e-05,9.208e-05
+10,3.68733,2.62766,3.67184,2.50334,0,0,0,0,1.44683,4.71852,2.19216,0.000102476,0.000102476,0.000102476
+11,3.97817,2.60241,4.1424,2.70991,0.00556,0.5,0.01605,0.0016,1.42181,4.70011,2.16441,0.000112625,0.000112625,0.000112625
+12,4.28608,3.28551,4.52089,2.67545,0.00588,0.5,0.01345,0.00269,1.52292,4.67835,2.13383,0.000122526,0.000122526,0.000122526
+13,4.58546,2.2033,4.2669,2.49512,0.00562,0.5,0.01157,0.00231,1.68387,4.61194,2.17338,0.00013218,0.00013218,0.00013218
+14,4.88485,2.87394,4.40826,2.70916,0.00568,0.5,0.00975,0.00195,1.89075,4.58147,2.30027,0.000141586,0.000141586,0.000141586
+15,5.19968,2.71982,4.27341,2.72914,0,0,0,0,1.96866,4.44855,2.33624,0.000150745,0.000150745,0.000150745
+16,5.49215,1.68605,4.24646,1.87531,0,0,0,0,2.00553,4.48707,2.29285,0.000159656,0.000159656,0.000159656
+17,5.79325,2.00138,3.94773,2.26758,0,0,0,0,2.09648,4.63246,2.4007,0.00016832,0.00016832,0.00016832
+18,6.09155,3.05671,5.53908,2.66707,0,0,0,0,2.09648,4.63246,2.4007,0.000176736,0.000176736,0.000176736
+19,6.39052,2.20901,3.52372,2.04393,0.00806,0.5,0.02369,0.00237,2.20865,4.54601,2.56285,0.000184905,0.000184905,0.000184905
+20,6.66892,2.4483,3.70065,2.23467,0.00806,0.5,0.02369,0.00237,2.20865,4.54601,2.56285,0.000192826,0.000192826,0.000192826
+21,6.9332,2.19448,3.86943,2.10117,0.00463,0.5,0.03317,0.00663,2.32559,4.48125,2.62548,0.0002005,0.0002005,0.0002005
+22,7.30519,2.9479,4.3599,2.76934,0.00463,0.5,0.03317,0.00663,2.32559,4.48125,2.62548,0.000207926,0.000207926,0.000207926
+23,7.68986,1.39714,3.40426,1.6151,0.01578,1,0.08837,0.01146,2.44041,4.6826,2.72339,0.000215105,0.000215105,0.000215105
+24,7.99967,2.46841,3.48691,2.13383,0.01578,1,0.08837,0.01146,2.44041,4.6826,2.72339,0.000222036,0.000222036,0.000222036
+25,8.31396,2.56835,4.10281,2.20972,0.00481,0.5,0.01777,0.00355,2.61255,4.61058,2.86581,0.00022872,0.00022872,0.00022872
+26,8.55453,1.65391,3.70866,1.88456,0.00481,0.5,0.01777,0.00355,2.61255,4.61058,2.86581,0.000235156,0.000235156,0.000235156
+27,8.80235,2.00119,3.45089,1.89926,0.01604,1,0.05212,0.00521,2.74287,4.6437,2.91424,0.000241345,0.000241345,0.000241345
+28,9.04528,1.89956,3.68555,2.16165,0.01604,1,0.05212,0.00521,2.74287,4.6437,2.91424,0.000247286,0.000247286,0.000247286
+29,9.29273,1.42472,2.92227,1.75125,0.01111,0.5,0.12437,0.01244,3.06977,4.5736,3.06093,0.00025298,0.00025298,0.00025298
+30,9.60359,1.35767,2.62078,1.55004,0.01111,0.5,0.12437,0.01244,3.06977,4.5736,3.06093,0.000258426,0.000258426,0.000258426
+31,9.94422,2.37035,4.66417,2.27559,0.01316,0.5,0.16583,0.01658,3.36893,4.24859,3.23425,0.000263625,0.000263625,0.000263625
+32,10.2594,1.61322,2.86179,1.53237,0.01316,0.5,0.16583,0.01658,3.36893,4.24859,3.23425,0.000268576,0.000268576,0.000268576
+33,10.6412,1.85856,3.15064,1.88045,0.02083,0.5,0.16583,0.01658,3.44544,4.47775,3.34489,0.00027328,0.00027328,0.00027328
+34,10.9661,1.62452,2.88937,1.72429,0.02083,0.5,0.16583,0.01658,3.44544,4.47775,3.34489,0.000277736,0.000277736,0.000277736
+35,11.2888,1.21431,3.51628,1.62717,0,0,0,0,3.47481,4.57555,3.33539,0.000281945,0.000281945,0.000281945
+36,11.5318,1.39337,2.94668,1.47319,0,0,0,0,3.47481,4.57555,3.33539,0.000285906,0.000285906,0.000285906
+37,11.7887,2.14196,3.0238,1.80533,0,0,0,0,3.7452,4.84862,3.44227,0.00028962,0.00028962,0.00028962
+38,12.0408,1.81846,2.33607,1.79617,0,0,0,0,3.7452,4.84862,3.44227,0.000293086,0.000293086,0.000293086
+39,12.2951,1.17175,1.92086,1.3652,0,0,0,0,4.06912,4.91101,3.62994,0.000296305,0.000296305,0.000296305
+40,12.5315,1.62593,2.48208,1.50529,0,0,0,0,4.06912,4.91101,3.62994,0.000299276,0.000299276,0.000299276
+41,12.8727,2.11165,2.67321,1.7814,0,0,0,0,4.39582,4.54093,3.81935,0.000302,0.000302,0.000302
+42,13.1287,1.58216,2.68408,1.63978,0,0,0,0,4.39582,4.54093,3.81935,0.000304476,0.000304476,0.000304476
+43,13.4027,1.4236,2.75829,1.527,0.01351,0.5,0.02163,0.00216,4.3829,4.34724,3.73475,0.000306705,0.000306705,0.000306705
+44,13.6611,1.58086,3.04931,1.97095,0.01351,0.5,0.02163,0.00216,4.3829,4.34724,3.73475,0.000308686,0.000308686,0.000308686
+45,13.94,1.70552,2.1028,1.7847,0.0119,0.5,0.03109,0.00311,4.02893,4.68709,3.46274,0.00031042,0.00031042,0.00031042
+46,14.2063,1.17287,2.30212,1.21687,0.0119,0.5,0.03109,0.00311,4.02893,4.68709,3.46274,0.000311906,0.000311906,0.000311906
+47,14.4787,0.75953,1.79745,0.99198,0.01111,0.5,0.04523,0.00452,4.16401,4.4644,3.55846,0.000313145,0.000313145,0.000313145
+48,14.7356,2.12559,4.16293,2.21951,0.01111,0.5,0.04523,0.00452,4.16401,4.4644,3.55846,0.000314136,0.000314136,0.000314136
+49,15.0753,0.92121,2.05049,1.10199,0.00962,0.5,0.04146,0.00829,3.90365,4.64333,3.48688,0.00031488,0.00031488,0.00031488
+50,15.3345,1.19831,1.99876,1.28056,0.00962,0.5,0.04146,0.00829,3.90365,4.64333,3.48688,0.000315376,0.000315376,0.000315376
+51,15.6062,2.135,2.40625,1.81904,0.01042,0.5,0.05528,0.0229,3.18088,5.77238,3.2612,0.000315625,0.000315625,0.000315625
+52,15.9429,1.72253,3.24847,1.71661,0.01042,0.5,0.05528,0.0229,3.18088,5.77238,3.2612,0.000315626,0.000315626,0.000315626
+53,16.2756,0.98827,2.52014,1.26921,0.01042,0.5,0.05528,0.0229,3.18088,5.77238,3.2612,0.00031538,0.00031538,0.00031538
+54,16.6349,1.8328,2.53368,1.66716,0.01316,0.5,0.12437,0.01244,3.8633,4.65938,3.23911,0.000314886,0.000314886,0.000314886
+55,16.8944,1.1818,1.78296,1.15008,0.01316,0.5,0.12437,0.01244,3.8633,4.65938,3.23911,0.000314145,0.000314145,0.000314145
+56,17.2351,1.26713,2.38825,1.40594,0.01316,0.5,0.12437,0.01244,3.8633,4.65938,3.23911,0.000313156,0.000313156,0.000313156
+57,17.5174,1.32827,1.94378,1.44297,0.01562,0.5,0.02073,0.00207,3.86529,4.77191,3.13775,0.00031192,0.00031192,0.00031192
+58,17.7998,0.93866,1.97455,1.19817,0.01562,0.5,0.02073,0.00207,3.86529,4.77191,3.13775,0.000310436,0.000310436,0.000310436
+59,18.0565,1.34164,4.69334,1.52493,0.01562,0.5,0.02073,0.00207,3.86529,4.77191,3.13775,0.000308705,0.000308705,0.000308705
+60,18.3262,1.70827,3.61104,1.67179,0,0,0,0,3.93656,4.99738,3.11248,0.000306726,0.000306726,0.000306726
+61,18.6133,0.82569,1.96032,1.1105,0,0,0,0,3.93656,4.99738,3.11248,0.0003045,0.0003045,0.0003045
+62,18.8669,1.33918,2.16787,1.37237,0,0,0,0,3.93656,4.99738,3.11248,0.000302026,0.000302026,0.000302026
+63,19.133,0.9404,2.00419,1.09295,0,0,0,0,3.54929,5.50927,3.04358,0.000299305,0.000299305,0.000299305
+64,19.4517,0.83957,1.71834,1.04536,0,0,0,0,3.54929,5.50927,3.04358,0.000296336,0.000296336,0.000296336
+65,19.7046,1.05975,1.98836,1.25573,0,0,0,0,3.54929,5.50927,3.04358,0.00029312,0.00029312,0.00029312
+66,19.966,0.79495,1.71511,1.07986,0.01111,0.5,0.02073,0.00415,3.48677,6.07562,2.94905,0.000289656,0.000289656,0.000289656
+67,20.2315,0.89674,1.76647,1.12805,0.01111,0.5,0.02073,0.00415,3.48677,6.07562,2.94905,0.000285945,0.000285945,0.000285945
+68,20.4951,1.27186,2.05522,1.24621,0.01111,0.5,0.02073,0.00415,3.48677,6.07562,2.94905,0.000281986,0.000281986,0.000281986
+69,20.763,0.72455,1.86789,1.0569,0,0,0,0,4.16796,5.98914,3.4865,0.00027778,0.00027778,0.00027778
+70,21.0206,0.94838,1.5672,1.15049,0,0,0,0,4.16796,5.98914,3.4865,0.000273326,0.000273326,0.000273326
+71,21.2798,1.62956,2.8345,1.75908,0,0,0,0,4.16796,5.98914,3.4865,0.000268625,0.000268625,0.000268625
+72,21.6185,0.82159,1.62683,0.94171,0,0,0,0,4.70051,5.39397,3.87655,0.000263676,0.000263676,0.000263676
+73,21.8885,0.84998,2.02611,1.30797,0,0,0,0,4.70051,5.39397,3.87655,0.00025848,0.00025848,0.00025848
+74,22.1454,0.85087,1.42472,1.15593,0,0,0,0,4.70051,5.39397,3.87655,0.000253036,0.000253036,0.000253036
+75,22.4095,1.1786,1.83474,1.21092,0,0,0,0,4.86689,6.24371,3.95955,0.000247345,0.000247345,0.000247345
+76,22.6694,0.78564,1.66187,0.97539,0,0,0,0,4.86689,6.24371,3.95955,0.000241406,0.000241406,0.000241406
+77,22.9286,0.79013,1.96926,1.26206,0,0,0,0,4.86689,6.24371,3.95955,0.00023522,0.00023522,0.00023522
+78,23.1943,0.8152,1.59465,1.07906,0,0,0,0,4.81344,5.97521,4.01022,0.000228786,0.000228786,0.000228786
+79,23.4546,0.61829,1.29337,0.98868,0,0,0,0,4.81344,5.97521,4.01022,0.000222105,0.000222105,0.000222105
+80,23.7685,0.95007,1.79471,1.1416,0,0,0,0,4.81344,5.97521,4.01022,0.000215176,0.000215176,0.000215176
+81,24.1034,0.93272,2.22583,1.20421,0.00617,0.5,0.00905,0.0009,4.60774,5.39816,3.81369,0.000208,0.000208,0.000208
+82,24.3629,0.63241,1.21512,1.08097,0.00617,0.5,0.00905,0.0009,4.60774,5.39816,3.81369,0.000200576,0.000200576,0.000200576
+83,24.6222,1.39649,2.39204,1.52958,0.00617,0.5,0.00905,0.0009,4.60774,5.39816,3.81369,0.000192905,0.000192905,0.000192905
+84,24.9101,0.63485,1.24139,0.96726,0.00641,0.5,0.01913,0.00459,4.14874,5.15654,3.60735,0.000184986,0.000184986,0.000184986
+85,25.1903,0.88995,1.76109,1.04753,0.00641,0.5,0.01913,0.00459,4.14874,5.15654,3.60735,0.00017682,0.00017682,0.00017682
+86,25.4597,1.16455,2.35159,1.15176,0.00641,0.5,0.01913,0.00459,4.14874,5.15654,3.60735,0.000168406,0.000168406,0.000168406
+87,25.761,0.86574,1.60094,1.1271,0.00641,0.5,0.01913,0.00459,4.14874,5.15654,3.60735,0.000159745,0.000159745,0.000159745
+88,26.1232,0.57931,1.17361,0.95885,0.00704,0.5,0.01658,0.00332,3.66565,5.88071,3.25055,0.000150836,0.000150836,0.000150836
+89,26.4022,0.88857,1.97356,1.07497,0.00704,0.5,0.01658,0.00332,3.66565,5.88071,3.25055,0.00014168,0.00014168,0.00014168
+90,26.7071,0.71015,1.22162,1.07789,0.00704,0.5,0.01658,0.00332,3.66565,5.88071,3.25055,0.000132276,0.000132276,0.000132276
+91,27.0615,0.80611,2.15097,0.95926,0.00704,0.5,0.01658,0.00332,3.66565,5.88071,3.25055,0.000122625,0.000122625,0.000122625
+92,27.3798,0.75395,1.9817,1.18844,0.00758,0.5,0.01463,0.00368,3.64756,5.57493,3.04703,0.000112726,0.000112726,0.000112726
+93,27.6734,0.52999,2.16673,0.94249,0.00758,0.5,0.01463,0.00368,3.64756,5.57493,3.04703,0.00010258,0.00010258,0.00010258
+94,27.9692,0.29519,1.34975,0.94508,0.00758,0.5,0.01463,0.00368,3.64756,5.57493,3.04703,9.21862e-05,9.21862e-05,9.21862e-05
+95,28.2525,0.533,1.80971,0.96369,0.00758,0.5,0.01463,0.00368,3.64756,5.57493,3.04703,8.1545e-05,8.1545e-05,8.1545e-05
+96,28.5584,0.37619,1.67301,0.82287,0.00714,0.5,0.01309,0.00361,3.45585,5.23022,2.909,7.06563e-05,7.06563e-05,7.06563e-05
+97,28.8522,0.58993,1.73032,0.93823,0.00714,0.5,0.01309,0.00361,3.45585,5.23022,2.909,5.952e-05,5.952e-05,5.952e-05
+98,29.2014,1.19689,2.95317,1.27242,0.00714,0.5,0.01309,0.00361,3.45585,5.23022,2.909,4.81363e-05,4.81363e-05,4.81363e-05
+99,29.5194,0.44641,1.50801,0.93925,0.00714,0.5,0.01309,0.00361,3.45585,5.23022,2.909,3.6505e-05,3.6505e-05,3.6505e-05
+100,29.8906,0.52149,1.81476,0.89577,0.0061,0.5,0.01157,0.00231,3.45621,5.51821,2.93443,2.46263e-05,2.46263e-05,2.46263e-05

ai/trainer.py

@@ -0,0 +1,269 @@
+# AI 학습 — YOLOv8 재학습 및 모델 저장
+import multiprocessing
+import os
+import random
+import shutil
+
+from utils.path_helper import get_path
+
+import yaml
+from PyQt5.QtCore import QThread, pyqtSignal
+
+
+class Trainer:
+    def __init__(self):
+        self.model       = None
+        self.is_training = False
+
+    # ------------------------------------------------------------------ #
+
+    def prepare_dataset(self, image_folder: str) -> str:
+        dataset_dir = get_path("ai", "dataset")
+        if os.path.exists(dataset_dir):
+            shutil.rmtree(dataset_dir)
+
+        for split in ("train", "val"):
+            os.makedirs(os.path.join(dataset_dir, "images", split), exist_ok=True)
+            os.makedirs(os.path.join(dataset_dir, "labels", split), exist_ok=True)
+
+        pairs = []
+        for f in os.listdir(image_folder):
+            if f.lower().endswith((".jpg", ".jpeg", ".png", ".bmp")):
+                img_path = os.path.join(image_folder, f)
+                txt_path = os.path.join(
+                    image_folder, os.path.splitext(f)[0] + ".txt"
+                )
+                if os.path.exists(txt_path):
+                    pairs.append((img_path, txt_path))
+
+        random.shuffle(pairs)
+        split_idx   = int(len(pairs) * 0.8)
+        train_pairs = pairs[:split_idx]
+        val_pairs   = pairs[split_idx:]
+
+        for img, lbl in train_pairs:
+            shutil.copy(img, os.path.join(dataset_dir, "images", "train"))
+            shutil.copy(lbl, os.path.join(dataset_dir, "labels", "train"))
+        for img, lbl in val_pairs:
+            shutil.copy(img, os.path.join(dataset_dir, "images", "val"))
+            shutil.copy(lbl, os.path.join(dataset_dir, "labels", "val"))
+
+        yaml_content = {
+            "path":  dataset_dir,
+            "train": "images/train",
+            "val":   "images/val",
+            "nc":    4,
+            "names": ["스크래치", "이물", "흑점", "변형"],
+        }
+        yaml_path = os.path.join(dataset_dir, "data.yaml")
+        with open(yaml_path, "w", encoding="utf-8") as fh:
+            yaml.dump(yaml_content, fh, allow_unicode=True)
+
+        return yaml_path
+
+    # ------------------------------------------------------------------ #
+
+    def train(
+        self,
+        image_folder:      str,
+        epochs:            int,
+        batch:             int,
+        save_path:         str,
+        log_callback=None,
+        progress_callback=None,
+    ):
+        self.is_training = True
+
+        try:
+            if log_callback:
+                log_callback("데이터셋 준비 중...")
+            yaml_path = self.prepare_dataset(image_folder)
+            if log_callback:
+                log_callback(f"데이터셋 준비 완료: {yaml_path}")
+
+            if log_callback:
+                log_callback("YOLOv8 모델 로드 중...")
+            from ultralytics import YOLO  # 지연 로딩 — 앱 시작 시 torch DLL 오류 방지
+            self.model = YOLO("yolov8n.pt")
+
+            if log_callback:
+                log_callback(f"학습 시작 (epoch={epochs}, batch={batch})")
+
+            def _on_epoch_end(trainer):
+                ep = trainer.epoch + 1
+                try:
+                    loss_val = float(trainer.loss)
+                    loss_str = f"{loss_val:.4f}"
+                except Exception:
+                    loss_str = "?"
+                if log_callback:
+                    log_callback(f"Epoch {ep}/{epochs}  loss={loss_str}")
+                if progress_callback:
+                    progress_callback(int(ep / epochs * 100))
+
+            self.model.add_callback("on_train_epoch_end", _on_epoch_end)
+
+            self.model.train(
+                data=yaml_path,
+                epochs=epochs,
+                batch=batch,
+                imgsz=640,
+                project=get_path("ai", "runs"),
+                name="train",
+                exist_ok=True,
+                verbose=True,
+                workers=0,    # disable DataLoader multiprocessing inside subprocess
+                amp=False,    # AMP check also spawns a subprocess on Windows
+                plots=False,  # matplotlib can interfere with Qt event loop
+            )
+
+            # best.pt 복사
+            best_pt = get_path("ai", "runs", "train", "weights", "best.pt")
+            if os.path.exists(best_pt):
+                os.makedirs(os.path.dirname(save_path), exist_ok=True)
+                shutil.copy(best_pt, save_path)
+                if log_callback:
+                    log_callback(f"모델 저장 완료: {save_path}")
+
+            if progress_callback:
+                progress_callback(100)
+            if log_callback:
+                log_callback("학습 완료!")
+
+        except BaseException as e:
+            import traceback
+            if log_callback:
+                try:
+                    log_callback(f"학습 오류: {e}")
+                    log_callback(traceback.format_exc())
+                except Exception:
+                    pass
+
+        finally:
+            self.is_training = False
+
+    # ------------------------------------------------------------------ #
+
+    def stop(self):
+        self.is_training = False
+
+
+# ====================================================================== #
+# Subprocess entry point — defined at module level so it is picklable.
+# ultralytics training can call sys.exit() internally; running it in a
+# separate process completely isolates the Qt application from that.
+# ====================================================================== #
+
+def _train_subprocess_main(queue, image_folder, epochs, batch, save_path):
+    """Entry point for the isolated training subprocess."""
+    try:
+        trainer = Trainer()
+
+        def _log(msg):
+            try:
+                queue.put(("log", msg))
+            except Exception:
+                pass
+
+        def _progress(pct):
+            try:
+                queue.put(("progress", int(pct)))
+            except Exception:
+                pass
+
+        trainer.train(
+            image_folder=image_folder,
+            epochs=epochs,
+            batch=batch,
+            save_path=save_path,
+            log_callback=_log,
+            progress_callback=_progress,
+        )
+        queue.put(("done", True))
+
+    except BaseException as e:
+        import traceback
+        try:
+            queue.put(("log", f"학습 오류: {e}"))
+            queue.put(("log", traceback.format_exc()))
+        except Exception:
+            pass
+        try:
+            queue.put(("done", False))
+        except Exception:
+            pass
+
+
+# ====================================================================== #
+
+
+class TrainWorker(QThread):
+    log_signal      = pyqtSignal(str)
+    progress_signal = pyqtSignal(int)
+    finished_signal = pyqtSignal(bool)
+
+    def __init__(self, trainer, image_folder, epochs, batch, save_path):
+        super().__init__()
+        self.trainer      = trainer
+        self.image_folder = image_folder
+        self.epochs       = epochs
+        self.batch        = batch
+        self.save_path    = save_path
+        self._proc        = None   # training subprocess handle
+
+    def run(self):
+        # Spawn an isolated subprocess so that any sys.exit() call inside
+        # ultralytics does not reach PyQt5's QThread handler and trigger
+        # QApplication.exit() in the main process.
+        ctx = multiprocessing.get_context("spawn")
+        q   = ctx.Queue()
+
+        self._proc = ctx.Process(
+            target=_train_subprocess_main,
+            args=(q, self.image_folder, self.epochs, self.batch, self.save_path),
+            daemon=True,
+        )
+        self._proc.start()
+
+        success = False
+        while True:
+            proc_alive = self._proc.is_alive()
+            try:
+                msg_type, msg_data = q.get(timeout=0.3)
+            except Exception:
+                # queue.Empty — check if subprocess died unexpectedly
+                if not proc_alive:
+                    # Give one last chance to read remaining messages
+                    while True:
+                        try:
+                            msg_type, msg_data = q.get_nowait()
+                        except Exception:
+                            break
+                        if msg_type == "log":
+                            self.log_signal.emit(str(msg_data))
+                        elif msg_type == "progress":
+                            self.progress_signal.emit(int(msg_data))
+                        elif msg_type == "done":
+                            success = bool(msg_data)
+                    break
+                continue
+
+            if msg_type == "log":
+                self.log_signal.emit(str(msg_data))
+            elif msg_type == "progress":
+                self.progress_signal.emit(int(msg_data))
+            elif msg_type == "done":
+                success = bool(msg_data)
+                break
+
+        self._proc.join(timeout=30)
+        if self._proc.is_alive():
+            self._proc.terminate()
+            self._proc.join(timeout=5)
+
+        self.finished_signal.emit(success)
+
+    def stop_subprocess(self):
+        """Call from the main thread to forcefully stop training."""
+        if self._proc and self._proc.is_alive():
+            self._proc.terminate()