add app.py + model files + requirements.txt

.ipynb_checkpoints/Untitled-checkpoint.ipynb

@@ -0,0 +1,130 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "f6a71737-070f-4eff-854b-b2432a032fdd",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch\n",
+    "import torch.nn as nn\n",
+    "import timm\n",
+    "import gradio as gr\n",
+    "from PIL import Image\n",
+    "\n",
+    "import torchvision.transforms as T\n",
+    "from ultralytics import YOLO\n",
+    "\n",
+    "device = torch.device(\"cpu\")\n",
+    "\n",
+    "num_classes = 3\n",
+    "classify_model = timm.create_model(\n",
+    "    \"efficientnet_b3\", pretrained=False, num_classes=num_classes\n",
+    ")\n",
+    "# 학습된 가중치 불러오기\n",
+    "state_dict = torch.load(\"classify.pth\", map_location=\"cpu\")\n",
+    "classify_model.load_state_dict(state_dict)\n",
+    "classify_model.to(device)\n",
+    "classify_model.eval()\n",
+    "\n",
+    "# 전처리 정의 (학습 때 사용한 방식과 동일해야 함)\n",
+    "transform_cls = T.Compose([\n",
+    "    T.Resize((224, 224)),\n",
+    "    T.ToTensor(),\n",
+    "    T.Normalize([0.485, 0.456, 0.406],\n",
+    "                [0.229, 0.224, 0.225]),\n",
+    "])\n",
+    "\n",
+    "yolo_model = torch.hub.load(\n",
+    "    \"ultralytics/yolov5\", \"yolov5n\", pretrained=False, source=\"github\"\n",
+    ")\n",
+    "yolo_model = torch.hub.load(\n",
+    "    \"ultralytics/yolov5\", \"custom\", path=\"detect.pt\", source=\"local\"\n",
+    ")\n",
+    "yolo_model.to(device)\n",
+    "yolo_model.eval()\n",
+    "\n",
+    "# 예측 함수\n",
+    "def pipeline_predict(img: Image.Image):\n",
+    "    # --- 분류 ---\n",
+    "    img_cls = transform_cls(img).unsqueeze(0).to(device)\n",
+    "    with torch.no_grad():\n",
+    "        out_cls = classify_model(img_cls)\n",
+    "        pred_cls = torch.argmax(out_cls, dim=1).item()\n",
+    "    cls_label = [\"정상\", \"경미 손상\", \"심각 손상\"][pred_cls]\n",
+    "\n",
+    "    # --- 객체 검출 ---\n",
+    "    results = yolo_model(img, size=640)\n",
+    "    det_img = results.render()[0]  # 시각화된 이미지\n",
+    "    det_img = Image.fromarray(det_img)\n",
+    "\n",
+    "    return cls_label, det_img"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "94211111-fadd-41dd-b812-7cd056ce5331",
+   "metadata": {},
+   "source": [
+    "# Gradio UI\n",
+    "demo = gr.Interface(\n",
+    "    fn=pipeline_predict,\n",
+    "    inputs=gr.Image(type=\"pil\"),\n",
+    "    outputs=[gr.Label(num_top_classes=1), gr.Image(type=\"pil\")],\n",
+    "    title=\"도로 파손 AI (경량화 버전)\",\n",
+    "    description=\"무료 HuggingFace Spaces CPU 환경에서도 동작하도록 최적화된 버전입니다.\",\n",
+    "    allow_flagging=\"never\"\n",
+    ")\n",
+    "\n",
+    "if __name__ == \"__main__\":\n",
+    "    demo.launch()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "265c26c0-4d58-4506-8a80-c059cfaed3d4",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "b13880fe-cae9-4215-b39d-c883466079d6",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ff17d6f2-7e3d-4c5c-8457-7ad197202c4a",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

.ipynb_checkpoints/Untitled1-checkpoint.ipynb

@@ -0,0 +1,154 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 42,
+   "id": "2617d8a8-b0f1-43a2-90c4-c0f0fc753b88",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch\n",
+    "import torch.nn.functional as F\n",
+    "import timm\n",
+    "import gradio as gr\n",
+    "from PIL import Image\n",
+    "import torchvision.transforms as T\n",
+    "from ultralytics import YOLO\n",
+    "import numpy as np\n",
+    "\n",
+    "device = torch.device(\"cpu\")\n",
+    "\n",
+    "# 분류 모델\n",
+    "num_classes = 3\n",
+    "classify_model = timm.create_model(\n",
+    "    \"efficientnet_b3\", pretrained=False, num_classes=num_classes)\n",
+    "\n",
+    "# 학습된 가중치 불러오기\n",
+    "state_dict = torch.load(\"classify.pth\", map_location=\"cpu\")\n",
+    "classify_model.load_state_dict(state_dict)\n",
+    "classify_model.to(device)\n",
+    "classify_model.eval()\n",
+    "\n",
+    "# 전처리\n",
+    "transform_cls = T.Compose([\n",
+    "    T.Resize((300, 300)),\n",
+    "    T.ToTensor(),\n",
+    "    T.Normalize([0.485, 0.456, 0.406],\n",
+    "                [0.229, 0.224, 0.225]),\n",
+    "])\n",
+    "\n",
+    "def fine_to_coarse(idx: int) -> str:\n",
+    "    return \"손상됨\" if idx in (1, 2) else \"정상\"\n",
+    "    \n",
+    "# 탐지 모델\n",
+    "yolo_model = YOLO(\"detect.pt\")   # 가중치 불러오기\n",
+    "yolo_model.to(device)\n",
+    "yolo_model.eval()\n",
+    "\n",
+    "# 예측 함수\n",
+    "def pipeline_predict(img: Image.Image):\n",
+    "    # 분류\n",
+    "    img_cls = transform_cls(img).unsqueeze(0).to(device)\n",
+    "    with torch.no_grad():\n",
+    "        out_cls = classify_model(img_cls)\n",
+    "        pred_cls = torch.argmax(out_cls, dim=1).item()\n",
+    "\n",
+    "    # 손상됨 상태면 탐지\n",
+    "    if pred_cls == 0:\n",
+    "        return \"정상\", img\n",
+    "    else:\n",
+    "        cls_label = \"손상됨\"\n",
+    "        # --- 객체 검출 (YOLOv8) ---\n",
+    "        results = yolo_model.predict(img, imgsz=640)\n",
+    "        det_img = results[0].plot()  # numpy (BGR)\n",
+    "        det_img = Image.fromarray(det_img[..., ::-1])\n",
+    "        return cls_label, det_img"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 43,
+   "id": "592edec4-f7ca-4b47-bcfd-0f203b397353",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "* Running on local URL:  http://127.0.0.1:7880\n",
+      "* To create a public link, set `share=True` in `launch()`.\n"
+     ]
+    },
+    {
+     "data": {
+      "text/html": [
+       "<div><iframe src=\"http://127.0.0.1:7880/\" width=\"100%\" height=\"500\" allow=\"autoplay; camera; microphone; clipboard-read; clipboard-write;\" frameborder=\"0\" allowfullscreen></iframe></div>"
+      ],
+      "text/plain": [
+       "<IPython.core.display.HTML object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "0: 640x640 1 십자파손, 189.5ms\n",
+      "Speed: 1.7ms preprocess, 189.5ms inference, 0.8ms postprocess per image at shape (1, 3, 640, 640)\n",
+      "\n",
+      "0: 640x640 1 횡방향균열, 191.7ms\n",
+      "Speed: 1.8ms preprocess, 191.7ms inference, 0.7ms postprocess per image at shape (1, 3, 640, 640)\n"
+     ]
+    }
+   ],
+   "source": [
+    "# ======================\n",
+    "# 5. Gradio UI\n",
+    "# ======================\n",
+    "demo = gr.Interface(\n",
+    "    fn=pipeline_predict,\n",
+    "    inputs=gr.Image(type=\"pil\"),\n",
+    "    outputs=[gr.Textbox(label=\"도로 상태\"), gr.Image(type=\"pil\", label=\"탐지 결과\")],\n",
+    "    title=\"도로 상태 분석\",\n",
+    "    description=\"분류 모델에서의 손상 확률이 기준 이상이면 탐지 모델로 파손 부분 탐지\",\n",
+    "    allow_flagging=\"never\"\n",
+    ")\n",
+    "\n",
+    "if __name__ == \"__main__\":\n",
+    "    demo.launch()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "692ac8db-ebb3-4a87-adec-151ac4459392",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

.ipynb_checkpoints/app-checkpoint.ipynb

@@ -0,0 +1,51 @@
+import torch
+import gradio as gr
+from PIL import Image
+import torchvision.transforms as transforms
+import yolov5  # ultralytics/yolov5 불러올 때 필요 (detect.pt 호환)
+
+# 1. 분류 모델 로드
+classify_model = torch.load("classify.pth", map_location="cpu")
+classify_model.eval()
+
+# 전처리 정의 (학습 때 사용한 방식과 동일해야 함)
+transform = transforms.Compose([
+    transforms.Resize((224,224)),
+    transforms.ToTensor(),
+    transforms.Normalize([0.485,0.456,0.406],[0.229,0.224,0.225])
+])
+
+# 2. 탐지 모델 로드
+detect_model = torch.hub.load("ultralytics/yolov5", "custom", path="detect.pt")
+
+# 예측 함수
+def pipeline(image):
+    # ---- 1단계 분류 ----
+    img = image.convert("RGB")
+    x = transform(img).unsqueeze(0)
+    with torch.no_grad():
+        out = classify_model(x)
+        pred = torch.argmax(out, dim=1).item()
+
+    # 분류 결과 해석 (예: 0=정상, 1=파손)
+    if pred == 0:
+        return f"도로 상태: 정상 (탐지 불필요)", image
+
+    # ---- 2단계 탐지 ----
+    results = detect_model(img)
+    detected_img = results.render()[0]  # numpy array
+    detected_img = Image.fromarray(detected_img)
+
+    return f"도로 상태: 파손 감지됨", detected_img
+
+# Gradio UI
+iface = gr.Interface(
+    fn=pipeline,
+    inputs=gr.Image(type="pil"),
+    outputs=[gr.Textbox(), gr.Image(type="pil")],
+    title="도로 상태 분석 파이프라인",
+    description="먼저 분류기로 도로 상태를 판별하고, 필요 시 탐지 모델로 파손 위치를 표시합니다."
+)
+
+if __name__ == "__main__":
+    iface.launch()

.ipynb_checkpoints/app-checkpoint.py

@@ -0,0 +1,51 @@
+import torch
+import gradio as gr
+from PIL import Image
+import torchvision.transforms as transforms
+import yolov5  # ultralytics/yolov5 불러올 때 필요 (detect.pt 호환)
+
+# 1. 분류 모델 로드
+classify_model = torch.load("classify.pth", map_location="cpu")
+classify_model.eval()
+
+# 전처리 정의 (학습 때 사용한 방식과 동일해야 함)
+transform = transforms.Compose([
+    transforms.Resize((224,224)),
+    transforms.ToTensor(),
+    transforms.Normalize([0.485,0.456,0.406],[0.229,0.224,0.225])
+])
+
+# 2. 탐지 모델 로드
+detect_model = torch.hub.load("ultralytics/yolov5", "custom", path="detect.pt")
+
+# 예측 함수
+def pipeline(image):
+    # ---- 1단계 분류 ----
+    img = image.convert("RGB")
+    x = transform(img).unsqueeze(0)
+    with torch.no_grad():
+        out = classify_model(x)
+        pred = torch.argmax(out, dim=1).item()
+
+    # 분류 결과 해석 (예: 0=정상, 1=파손)
+    if pred == 0:
+        return f"도로 상태: 정상 (탐지 불필요)", image
+
+    # ---- 2단계 탐지 ----
+    results = detect_model(img)
+    detected_img = results.render()[0]  # numpy array
+    detected_img = Image.fromarray(detected_img)
+
+    return f"도로 상태: 파손 감지됨", detected_img
+
+# Gradio UI
+iface = gr.Interface(
+    fn=pipeline,
+    inputs=gr.Image(type="pil"),
+    outputs=[gr.Textbox(), gr.Image(type="pil")],
+    title="도로 상태 분석 파이프라인",
+    description="먼저 분류기로 도로 상태를 판별하고, 필요 시 탐지 모델로 파손 위치를 표시합니다."
+)
+
+if __name__ == "__main__":
+    iface.launch()

Untitled.ipynb

@@ -0,0 +1,122 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "f6a71737-070f-4eff-854b-b2432a032fdd",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Downloading: \"https://github.com/ultralytics/yolov5/zipball/master\" to /home/user22313548/.cache/torch/hub/master.zip\n"
+     ]
+    }
+   ],
+   "source": [
+    "import torch\n",
+    "import torch.nn as nn\n",
+    "import timm\n",
+    "import gradio as gr\n",
+    "from PIL import Image\n",
+    "\n",
+    "import torchvision.transforms as T\n",
+    "from ultralytics import YOLO\n",
+    "\n",
+    "device = torch.device(\"cpu\")\n",
+    "\n",
+    "num_classes = 3\n",
+    "classify_model = timm.create_model(\n",
+    "    \"efficientnet_b3\", pretrained=False, num_classes=num_classes\n",
+    ")\n",
+    "# 학습된 가중치 불러오기\n",
+    "state_dict = torch.load(\"classify.pth\", map_location=\"cpu\")\n",
+    "classify_model.load_state_dict(state_dict)\n",
+    "classify_model.to(device)\n",
+    "classify_model.eval()\n",
+    "\n",
+    "# 전처리 정의 (학습 때 사용한 방식과 동일해야 함)\n",
+    "transform_cls = T.Compose([\n",
+    "    T.Resize((224, 224)),\n",
+    "    T.ToTensor(),\n",
+    "    T.Normalize([0.485, 0.456, 0.406],\n",
+    "                [0.229, 0.224, 0.225]),\n",
+    "])\n",
+    "\n",
+    "yolo_model = torch.hub.load(\n",
+    "    \"ultralytics/yolov5\", \"yolov5n\", pretrained=False, source=\"github\"\n",
+    ")\n",
+    "yolo_model = torch.hub.load(\n",
+    "    \"ultralytics/yolov5\", \"custom\", path=\"detect.pt\", source=\"local\"\n",
+    ")\n",
+    "yolo_model.to(device)\n",
+    "yolo_model.eval()\n",
+    "\n",
+    "# 예측 함수\n",
+    "def pipeline_predict(img: Image.Image):\n",
+    "    # --- 분류 ---\n",
+    "    img_cls = transform_cls(img).unsqueeze(0).to(device)\n",
+    "    with torch.no_grad():\n",
+    "        out_cls = classify_model(img_cls)\n",
+    "        pred_cls = torch.argmax(out_cls, dim=1).item()\n",
+    "    cls_label = [\"정상\", \"경미 손상\", \"심각 손상\"][pred_cls]\n",
+    "\n",
+    "    # --- 객체 검출 ---\n",
+    "    results = yolo_model(img, size=640)\n",
+    "    det_img = results.render()[0]  # 시각화된 이미지\n",
+    "    det_img = Image.fromarray(det_img)\n",
+    "\n",
+    "    return cls_label, det_img"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "94211111-fadd-41dd-b812-7cd056ce5331",
+   "metadata": {},
+   "source": [
+    "# Gradio UI\n",
+    "demo = gr.Interface(\n",
+    "    fn=pipeline_predict,\n",
+    "    inputs=gr.Image(type=\"pil\"),\n",
+    "    outputs=[gr.Label(num_top_classes=1), gr.Image(type=\"pil\")],\n",
+    "    title=\"도로 파손 AI (경량화 버전)\",\n",
+    "    description=\"무료 HuggingFace Spaces CPU 환경에서도 동작하도록 최적화된 버전입니다.\",\n",
+    "    allow_flagging=\"never\"\n",
+    ")\n",
+    "\n",
+    "if __name__ == \"__main__\":\n",
+    "    demo.launch()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "239606a5-63df-49f6-a1dd-69ba979a3e21",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

Untitled1.ipynb

@@ -0,0 +1,151 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 44,
+   "id": "2617d8a8-b0f1-43a2-90c4-c0f0fc753b88",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch\n",
+    "import torch.nn.functional as F\n",
+    "import timm\n",
+    "import gradio as gr\n",
+    "from PIL import Image\n",
+    "import torchvision.transforms as T\n",
+    "from ultralytics import YOLO\n",
+    "import numpy as np\n",
+    "\n",
+    "device = torch.device(\"cpu\")\n",
+    "\n",
+    "# 분류 모델\n",
+    "num_classes = 3\n",
+    "classify_model = timm.create_model(\n",
+    "    \"efficientnet_b3\", pretrained=False, num_classes=num_classes)\n",
+    "\n",
+    "# 학습된 가중치 불러오기\n",
+    "state_dict = torch.load(\"classify.pth\", map_location=\"cpu\")\n",
+    "classify_model.load_state_dict(state_dict)\n",
+    "classify_model.to(device)\n",
+    "classify_model.eval()\n",
+    "\n",
+    "# 전처리\n",
+    "transform_cls = T.Compose([\n",
+    "    T.Resize((300, 300)),\n",
+    "    T.ToTensor(),\n",
+    "    T.Normalize([0.485, 0.456, 0.406],\n",
+    "                [0.229, 0.224, 0.225]),\n",
+    "])\n",
+    "\n",
+    "def fine_to_coarse(idx: int) -> str:\n",
+    "    return \"손상됨\" if idx in (1, 2) else \"정상\"\n",
+    "    \n",
+    "# 탐지 모델\n",
+    "yolo_model = YOLO(\"detect.pt\")   # 가중치 불러오기\n",
+    "yolo_model.to(device)\n",
+    "yolo_model.eval()\n",
+    "\n",
+    "# 예측 함수\n",
+    "def pipeline_predict(img: Image.Image):\n",
+    "    # 분류\n",
+    "    img_cls = transform_cls(img).unsqueeze(0).to(device)\n",
+    "    with torch.no_grad():\n",
+    "        out_cls = classify_model(img_cls)\n",
+    "        pred_cls = torch.argmax(out_cls, dim=1).item()\n",
+    "\n",
+    "    # 손상됨 상태면 탐지\n",
+    "    if pred_cls == 0:\n",
+    "        return \"정상\", img\n",
+    "    else:\n",
+    "        cls_label = \"손상됨\"\n",
+    "        # --- 객체 검출 (YOLOv8) ---\n",
+    "        results = yolo_model.predict(img, imgsz=640)\n",
+    "        det_img = results[0].plot()  # numpy (BGR)\n",
+    "        det_img = Image.fromarray(det_img[..., ::-1])\n",
+    "        return cls_label, det_img"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 45,
+   "id": "592edec4-f7ca-4b47-bcfd-0f203b397353",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "* Running on local URL:  http://127.0.0.1:7881\n",
+      "* To create a public link, set `share=True` in `launch()`.\n"
+     ]
+    },
+    {
+     "data": {
+      "text/html": [
+       "<div><iframe src=\"http://127.0.0.1:7881/\" width=\"100%\" height=\"500\" allow=\"autoplay; camera; microphone; clipboard-read; clipboard-write;\" frameborder=\"0\" allowfullscreen></iframe></div>"
+      ],
+      "text/plain": [
+       "<IPython.core.display.HTML object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "0: 640x640 1 십자파손, 193.6ms\n",
+      "Speed: 1.7ms preprocess, 193.6ms inference, 0.8ms postprocess per image at shape (1, 3, 640, 640)\n"
+     ]
+    }
+   ],
+   "source": [
+    "# ======================\n",
+    "# 5. Gradio UI\n",
+    "# ======================\n",
+    "demo = gr.Interface(\n",
+    "    fn=pipeline_predict,\n",
+    "    inputs=gr.Image(type=\"pil\"),\n",
+    "    outputs=[gr.Textbox(label=\"도로 상태\"), gr.Image(type=\"pil\", label=\"탐지 결과\")],\n",
+    "    title=\"도로 상태 분석\",\n",
+    "    description=\"분류 모델에서의 손상 확률이 기준 이상이면 탐지 모델로 파손 부분 탐지\",\n",
+    "    allow_flagging=\"never\"\n",
+    ")\n",
+    "\n",
+    "if __name__ == \"__main__\":\n",
+    "    demo.launch()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "692ac8db-ebb3-4a87-adec-151ac4459392",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

app.py

@@ -0,0 +1,69 @@
+import torch
+import torch.nn.functional as F
+import timm
+import gradio as gr
+from PIL import Image
+import torchvision.transforms as T
+from ultralytics import YOLO
+import numpy as np
+
+device = torch.device("cpu")
+
+# 분류 모델
+num_classes = 3
+classify_model = timm.create_model(
+    "efficientnet_b3", pretrained=False, num_classes=num_classes)
+
+# 학습된 가중치 불러오기
+state_dict = torch.load("classify.pth", map_location="cpu")
+classify_model.load_state_dict(state_dict)
+classify_model.to(device)
+classify_model.eval()
+
+# 전처리
+transform_cls = T.Compose([
+    T.Resize((300, 300)),
+    T.ToTensor(),
+    T.Normalize([0.485, 0.456, 0.406],
+                [0.229, 0.224, 0.225]),
+])
+
+def fine_to_coarse(idx: int) -> str:
+    return "손상됨" if idx in (1, 2) else "정상"
+    
+# 탐지 모델
+yolo_model = YOLO("detect.pt")   # 가중치 불러오기
+yolo_model.to(device)
+yolo_model.eval()
+
+# 예측 함수
+def pipeline_predict(img: Image.Image):
+    # 분류
+    img_cls = transform_cls(img).unsqueeze(0).to(device)
+    with torch.no_grad():
+        out_cls = classify_model(img_cls)
+        pred_cls = torch.argmax(out_cls, dim=1).item()
+
+    # 손상됨 상태면 탐지
+    if pred_cls == 0:
+        return "정상", img
+    else:
+        cls_label = "손상됨"
+        # --- 객체 검출 (YOLOv8) ---
+        results = yolo_model.predict(img, imgsz=640)
+        det_img = results[0].plot()  # numpy (BGR)
+        det_img = Image.fromarray(det_img[..., ::-1])
+        return cls_label, det_img
+
+# Gradio UI
+demo = gr.Interface(
+    fn=pipeline_predict,
+    inputs=gr.Image(type="pil"),
+    outputs=[gr.Textbox(label="도로 상태"), gr.Image(type="pil", label="탐지 결과")],
+    title="도로 상태 분석",
+    description="분류 모델에서의 손상 확률이 기준 이상이면 탐지 모델로 파손 부분 탐지",
+    allow_flagging="never"
+)
+
+if __name__ == "__main__":
+    demo.launch()

classify.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74ff4565abff0fd34ad6daf409a073774b55601554837131e50b9281dfd1366b
+size 43360699

detect.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d9d8607b52af8891335957573247fa8305462f9c8f4c10e033729c66906c837
+size 52049362

requirements.txt

@@ -0,0 +1,7 @@
+torch
+torchvision
+timm
+gradio
+Pillow
+ultralytics
+numpy