Update app.py

app.py

@@ -1,110 +1,105 @@
-import gradio as gr
-import numpy as np
-from PIL import Image
-import tensorflow as tf
-
-# ------------------------------
-# 1) 모델 파일 경로
-# ------------------------------
-MODEL_PATH = "crack_detection.h5"
-
-# ------------------------------
-# 2) TensorFlow 모델 로드
-# ------------------------------
-model = tf.keras.models.load_model(MODEL_PATH)
-print("🔥 Loaded TensorFlow crack classifier")
-print("   Input shape :", model.input_shape)
-print("   Output shape:", model.output_shape)
-
-# (None, H, W, C) 형태라고 가정
-input_shape = model.input_shape
-if len(input_shape) != 4:
-    raise ValueError(f"예상치 못한 input_shape: {input_shape}")
-_, H, W, C = input_shape
-
-
-# ------------------------------
-# 3) 예측 함수
-#    항상 JSON을 리턴하도록 try/except
-# ------------------------------
-def predict(img: Image.Image):
-    try:
-        # 1) 이미지 전처리
-        img = img.convert("RGB")
-        img_resized = img.resize((W, H))
-        arr = np.array(img_resized).astype("float32") / 255.0
-        X = np.expand_dims(arr, axis=0)  # (1, H, W, C)
-
-        # 2) 모델 추론
-        raw = model.predict(X)[0]
-        probs = np.array(raw, dtype="float32").flatten()
-
-        # 3) 출력 해석
-        #    - 길이 1  : sigmoid → p_crack
-        #    - 길이 2+ : [p_normal, p_crack] 가정
-        if probs.shape[0] == 1:
-            p_crack = float(probs[0])
-            p_normal = 1.0 - p_crack
-
-            if p_crack >= 0.5:
-                label = "crack"
-                conf = p_crack
-            else:
-                label = "normal"
-                conf = p_normal
-
-        elif probs.shape[0] >= 2:
-            p_normal = float(probs[0])
-            p_crack = float(probs[1])
-
-            if p_crack >= p_normal:
-                label = "crack"
-                conf = p_crack
-            else:
-                label = "normal"
-                conf = p_normal
-        else:
-            # 매우 특이한 케이스 → 그냥 argmax
-            idx = int(np.argmax(probs))
-            label = f"class_{idx}"
-            conf = float(probs[idx])
-
-        return {
-            "data": [
-                {
-                    "label": label,
-                    "confidence": float(conf),
-                }
-            ]
-        }
-
-    except Exception as e:
-        # ❗ 여기서 예외를 모두 잡아서 JSON으로 내려줌
-        #    이렇게 해야 HF Space가 500 안 던지고,
-        #    프론트에서 Raw Response/JSON Payload를 볼 수 있음.
-        print("❌ Error in predict():", repr(e))
-        return {
-            "data": [
-                {
-                    "label": "error",
-                    "confidence": 0.0,
-                    "message": str(e),
-                }
-            ]
-        }
-
-
-# ------------------------------
-# 4) Gradio Interface
-# ------------------------------
-demo = gr.Interface(
-    fn=predict,
-    inputs=gr.Image(type="pil", label="Input image"),
-    outputs=gr.JSON(label="Detection Result"),
-    title="Crack Detection Classifier (Keras .h5)",
-    description="사진을 업로드하면 균열/정상 여부와 확률(%)을 반환합니다.",
-    flagging_mode="never",
-)
-
-if __name__ == "__main__":
-    demo.launch()
+/// <reference types="vite/client" />
+
+import { client } from "@gradio/client";
+
+export const HF_SPACE_URL = "https://thompson001-crack-api.hf.space/";
+
+export interface InferenceResult {
+  hasCrack: boolean;
+  confidence: number;
+  label: string;
+  details: string;
+  isCustomModel: boolean;
+}
+
+// Gradio Result 구조
+interface GradioEvent {
+  data: any[];
+  [key: string]: any;
+}
+
+interface HFJsonItem {
+  label?: string;
+  confidence?: number;
+  message?: string;
+}
+
+interface HFJsonPayload {
+  data: HFJsonItem[];
+}
+
+// -------------------------------------------------------
+// HF Space 호출
+// -------------------------------------------------------
+export async function predictCrack(file: File): Promise<InferenceResult> {
+  console.log("🔵 Connecting to HF Space…");
+
+  let app;
+  try {
+    app = await client(HF_SPACE_URL);
+  } catch (err) {
+    console.error("❌ HF Space 연결 실패:", err);
+    throw new Error("HF Space 연결 실패");
+  }
+
+  console.log("🔵 Sending image to /predict…");
+
+  let rawResult: GradioEvent;
+  try {
+    rawResult = (await app.predict("/predict", {
+      img: file, // 반드시 img
+    })) as GradioEvent;
+  } catch (err: any) {
+    console.error("❌ /predict 호출 실패:", err);
+    throw new Error("HF Space 예측 호출 실패");
+  }
+
+  console.log("🔵 Raw Response:", rawResult);
+
+  const jsonPayload = rawResult?.data?.[0] as HFJsonPayload;
+
+  console.log("🔵 JSON Payload:", jsonPayload);
+
+  // JSON 없는 경우 → 정상 취급
+  if (!jsonPayload?.data || jsonPayload.data.length === 0) {
+    return {
+      hasCrack: false,
+      confidence: 1,
+      label: "정상",
+      details: "",
+      isCustomModel: true,
+    };
+  }
+
+  const first: HFJsonItem = jsonPayload.data[0];
+
+  const rawLabel = typeof first.label === "string" ? first.label : "normal";
+  const confidence =
+    typeof first.confidence === "number" ? first.confidence : 1;
+
+  const normalized = rawLabel.replace(/\s+/g, "").toLowerCase();
+
+  // 에러 처리
+  if (normalized === "error") {
+    throw new Error(first.message || "모델 내부 오류");
+  }
+
+  // 균열 판단: crack, damaged 등 포함
+  const isCrack =
+    normalized.includes("crack") ||
+    normalized.includes("positive") ||
+    normalized.includes("damaged");
+
+  // 🔥 너가 원한 로직: 정상일 때는 % 숨김
+  const details = isCrack
+    ? `균열 확률 ${(confidence * 100).toFixed(1)}%`
+    : "";
+
+  return {
+    hasCrack: isCrack,
+    confidence,
+    label: isCrack ? "균열 감지" : "정상",
+    details,
+    isCustomModel: true,
+  };
+}