feat: Enable direct file uploads for face verification and quality validation, and document free-tier infrastructure considerations.

app.py

@@ -9,16 +9,12 @@ from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel, HttpUrl
 from inference import FaceAnalysis
 
-# ==========================================
-# INISIALISASI APLIKASI & MODEL
-# ==========================================
 app = FastAPI(
     title="GawayKu Face Recognition API",
     description="API Face Recognition mandiri menggunakan YOLOv8 + ArcFace untuk aplikasi HRIS GawayKu.",
-    version="1.1.0",
+    version="1.2.0",
 )
 
-# CORS — agar bisa diakses dari frontend mana saja
 app.add_middleware(
     CORSMiddleware,
     allow_origins=["*"],
@@ -27,11 +23,9 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
-# Folder sementara untuk menyimpan file upload
 TEMP_DIR = "/tmp/face_uploads"
 os.makedirs(TEMP_DIR, exist_ok=True)
 
-# Inisialisasi model saat server pertama kali menyala
 print("⏳ Memuat model Face Recognition...")
 try:
     face_app = FaceAnalysis()
@@ -41,15 +35,9 @@ except Exception as e:
     face_app = None
 
 # ==========================================
-# MODELS (Pydantic)
+# MODELS
 # ==========================================
-
-class VerifyRequest(BaseModel):
-    reference_image_url: HttpUrl
-    query_image_url: HttpUrl
-
-class ValidateQualityRequest(BaseModel):
-    image_url: HttpUrl
+# (VerifyRequest biasa dihapus karena kita pakai Form Data + UploadFile)
 
 class VerifyResponse(BaseModel):
     is_match: bool
@@ -66,25 +54,19 @@ class ValidationResponse(BaseModel):
 # ==========================================
 # HELPER FUNCTIONS
 # ==========================================
-
 def download_image(url: str) -> str:
-    """Mendownload gambar dari URL dan menyimpannya sementara."""
     try:
         response = requests.get(url, timeout=10)
         response.raise_for_status()
-        
         unique_name = f"{uuid.uuid4().hex}.jpg"
         file_path = os.path.join(TEMP_DIR, unique_name)
-        
         with open(file_path, "wb") as f:
             f.write(response.content)
-            
         return file_path
     except Exception as e:
-        raise HTTPException(status_code=400, detail=f"Gagal mendownload gambar: {str(e)}")
+        raise HTTPException(status_code=400, detail=f"Gagal mendownload gambar referensi: {str(e)}")
 
 def clear_temp_file(path: str):
-    """Menghapus file sementara."""
     if path and os.path.exists(path):
         os.remove(path)
 
@@ -94,39 +76,40 @@ def clear_temp_file(path: str):
 
 @app.get("/")
 def health_check():
-    """Health check endpoint — memastikan server berjalan."""
     status = "ok" if face_app else "error"
     return {
         "status": status,
-        "message": "🚀 GawayKu Face Recognition API is running!",
+        "message": "🚀 GawayKu Face Recognition API is running (v1.2 - Direct Upload)!",
         "model": "YOLOv8 + ArcFace (WideResNet-101-2)",
     }
 
-@app.post("/verify", response_model=VerifyResponse)
-async def verify_faces(request: VerifyRequest):
+@app.post("/verify-file", response_model=VerifyResponse)
+async def verify_file(
+    reference_image_url: str = Form(...),
+    query_image: UploadFile = File(...)
+):
     """
-    Membandingkan dua gambar wajah dari URL.
+    Membandingkan Wajah Referensi (URL) vs Wajah Query (File Upload Langsung).
     
-    Returns:
-        - is_match: bool (True jika wajah sama)
-        - similarity_score: float (0.0 - 1.0)
-        - execution_time: float (detik)
-        - error: str (pesan error jika ada)
+    - reference_image_url: URL foto wajah yang sudah terdaftar di database.
+    - query_image: File foto selfie langsung dari mobile (multipart/form-data).
     """
     if not face_app:
         raise HTTPException(status_code=503, detail="Model belum siap")
 
     start_time = time.time()
-    path1 = None
-    path2 = None
+    path_ref = None
 
     try:
-        # Download gambar
-        path1 = download_image(str(request.reference_image_url))
-        path2 = download_image(str(request.query_image_url))
+        # 1. Download Reference Image (URL)
+        path_ref = download_image(reference_image_url)
 
-        # Proses perbandingan wajah
-        similarity, is_same = face_app.compare(path1, path2)
+        # 2. Read Query Image (Bytes dari UploadFile)
+        query_bytes = await query_image.read()
+
+        # 3. Bandingkan (Path vs Bytes)
+        # Kita update process_image di inference.py untuk terima bytes juga
+        similarity, is_same = face_app.compare(path_ref, query_bytes)
         
         exec_time = time.time() - start_time
 
@@ -146,36 +129,24 @@ async def verify_faces(request: VerifyRequest):
             "error": str(e)
         }
     finally:
-        clear_temp_file(path1)
-        clear_temp_file(path2)
+        clear_temp_file(path_ref)
+        # query_bytes ada di memory, tidak perlu delete file
 
-@app.post("/validate-quality", response_model=ValidationResponse)
-async def validate_quality(request: ValidateQualityRequest):
+@app.post("/validate-quality-file", response_model=ValidationResponse)
+async def validate_quality_file(image: UploadFile = File(...)):
     """
-    Memeriksa kualitas foto untuk pendaftaran (Enrollment).
-    Memastikan wajah terdeteksi dengan jelas dan tunggal.
+    Cek kualitas wajah dari file upload langsung (tanpa URL).
     """
     if not face_app:
         raise HTTPException(status_code=503, detail="Model belum siap")
 
-    path = None
     try:
-        path = download_image(str(request.image_url))
-        
-        # Gunakan method internal process_image untuk mendapatkan embedding
-        # Jika berhasil return embedding, artinya wajah terdeteksi
-        # TODO: Idealnya face_app.process_image melempar error specifik jika wajah tak ditemukan
-        # Saat ini kita asumsikan jika berhasil embedding = valid
-        
-        embedding = face_app.process_image(path)
-        
-        # Simulasi cek kualitas (Placeholder)
-        # Di implementasi nyata, kita bisa cek resolusi crop wajah, blur, dll.
-        # Karena process_image mengembalikan embedding, kita anggap valid.
+        image_bytes = await image.read()
+        embedding = face_app.process_image(image_bytes)
         
         return {
             "is_valid": True,
-            "quality_score": 0.95, # Dummy score
+            "quality_score": 0.95,
             "message": "Wajah terdeteksi dengan baik.",
             "error": None
         }
@@ -186,6 +157,4 @@ async def validate_quality(request: ValidateQualityRequest):
             "quality_score": 0.0,
             "message": "Gagal memproses wajah.",
             "error": str(e)
-        }
-    finally:
-        clear_temp_file(path)
+        }

attendance_system_context.md

@@ -11,7 +11,26 @@ Ekosistem GawayKu terdiri dari:
 2.  **Backend WebAdmin (Next.js):** Mengelola pengguna, menyimpan data wajah referensi, dan log absensi.
 3.  **Layanan Face Recognition (Python/FastAPI):** Layanan terpisah (yang akan dibuat) untuk verifikasi biometrik.
 
-## 2. Alur Kerja Face Recognition
+## 2. Pertimbangan Infrastruktur (Free Tier)
+
+Karena menggunakan **Vercel (Free)**, **Supabase (Free)**, dan **Hugging Face Spaces (Free)**, berikut adalah batasan dan solusi:
+
+### **Vercel (Backend API)**
+-   **Aman:** Vercel hanya meneruskan request ke Hugging Face. Beban komputasi ada di Hugging Face.
+-   **Risiko:** Vercel Function (Free) memiliki timeout 10-60 detik. Jika Hugging Face sedang *cold start* (tidur), request dari Vercel bisa timeout.
+-   **Solusi:** Pastikan Hugging Face selalu aktif (lihat poin Hugging Face).
+
+### **Supabase (Database) & Cloudinary (Storage)**
+-   **Database (Supabase):** Aman (URL text sangat ringan).
+-   **Storage (Cloudinary):** Jauh lebih baik! Free tier Cloudinary (~25 Credits) sangat cukup untuk ribuan foto absensi.
+-   **Keuntungan:** Cloudinary otomatis mengoptimalkan delivery gambar, jadi download ke API Face Recognition bisa lebih cepat.
+-   **Solusi:** Tetap lakukan resize di mobile app untuk menghemat kuota bandwidth Cloudinary.
+
+### **Hugging Face Spaces (CPU Basic)**
+-   **Isu Utama:** Space akan "tidur" setelah 48 jam tidak aktif. *Cold start* butuh 2-3 menit.
+-   **Solusi (Wajib):** Setup cron-job gratis (misal: console.cron-job.org) untuk "ping" endpoint Health Check (`GET /`) setiap 1 jam agar Space tidak pernah tidur.
+
+## 3. Alur Kerja Face Recognition
 
 Sistem ini memiliki dua fase utama: **Pendaftaran (Enrollment)** dan **Verifikasi (Verification)**.