import os from io import BytesIO from pathlib import Path from flask import Flask, jsonify, request from huggingface_hub import hf_hub_download import csv from datetime import datetime import cv2 import torch import torch.nn as nn import numpy as np from torchvision import models, transforms from PIL import Image, ImageOps app = Flask(__name__) # ============================================================ # HOME # ============================================================ @app.route("/") def home(): return jsonify({ "status": "success", "message": "Backend is running" }), 200 # ============================================================ # HEALTH # ============================================================ @app.route("/health", methods=["GET", "POST"]) def health(): return jsonify({ "status": "healthy" }), 200 # ============================================================ # MODEL 0 # ============================================================ @app.route("/model/0", methods=["GET", "POST"]) def model_0(): # ======================================================== # GET REQUEST # ======================================================== if request.method == "GET": return jsonify({ "status": "model is running", "model": "MobileNetV3-Small", "classes": [ "Sajan", "UNKNOWN" ] }), 200 try: # ==================================================== # CONFIG # ==================================================== SAJAN_THRESHOLD = 0.80 device = torch.device("cpu") CSV_FILE = "predictions.csv" # ==================================================== # DOWNLOAD HOSTED MODEL # ==================================================== MODEL_PATH = hf_hub_download( repo_id="ajaycodeself/ml_models", filename="models/model0.pth", repo_type="space" ) # ==================================================== # LOAD CHECKPOINT # ==================================================== checkpoint = torch.load( MODEL_PATH, map_location=device, weights_only=False ) classes = checkpoint["classes"] # ==================================================== # CREATE MOBILENETV3-SMALL # ==================================================== model = models.mobilenet_v3_small(weights=None) model.classifier[3] = nn.Linear( model.classifier[3].in_features, len(classes) ) # ==================================================== # LOAD TRAINED WEIGHTS # ==================================================== model.load_state_dict(checkpoint["model_state_dict"]) model.to(device) model.eval() # ==================================================== # PREPROCESSING # ==================================================== transform = transforms.Compose([ transforms.ToPILImage(), transforms.Resize((224, 224)), transforms.ToTensor(), transforms.Normalize( mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225] ) ]) # ==================================================== # HAAR FACE DETECTOR # ==================================================== face_detector = cv2.CascadeClassifier( cv2.data.haarcascades + "haarcascade_frontalface_default.xml" ) # ==================================================== # CHECK IMAGE # ==================================================== if "image" not in request.files: return jsonify({ "status": "error", "message": "image is required" }), 400 # ==================================================== # READ IMAGE # ==================================================== file = request.files["image"] image_bytes = file.read() image_array = np.frombuffer(image_bytes, dtype=np.uint8) frame = cv2.imdecode(image_array, cv2.IMREAD_COLOR) if frame is None: return jsonify({ "status": "error", "message": "invalid image" }), 400 # ==================================================== # FACE DETECTION # ==================================================== gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY) faces = face_detector.detectMultiScale( gray, scaleFactor=1.1, minNeighbors=6, minSize=(50, 50) ) # ==================================================== # NO FACE = UNKNOWN # ==================================================== if len(faces) == 0: return jsonify({ "status": "success", "prediction": "UNKNOWN", "confidence": 0, "confidence_percent": 0, "face_detected": False, "bounding_box": None }), 200 # ==================================================== # LARGEST FACE # ==================================================== x, y, w, h = max( faces, key=lambda face: face[2] * face[3] ) # ==================================================== # FACE MARGIN # ==================================================== margin_x = int(w * 0.15) margin_y = int(h * 0.15) x1 = max(0, x - margin_x) y1 = max(0, y - margin_y) x2 = min(frame.shape[1], x + w + margin_x) y2 = min(frame.shape[0], y + h + margin_y) # ==================================================== # CROP FACE # ==================================================== face = frame[y1:y2, x1:x2] if face.size == 0: return jsonify({ "status": "success", "prediction": "UNKNOWN", "confidence": 0, "confidence_percent": 0, "face_detected": False, "bounding_box": None }), 200 # ==================================================== # PREPROCESS FACE # ==================================================== tensor = transform(face).unsqueeze(0).to(device) # ==================================================== # MODEL PREDICTION # ==================================================== with torch.inference_mode(): output = model(tensor) probabilities = torch.softmax(output, dim=1) # ==================================================== # FIND SAJAN CLASS # ==================================================== sajan_index = None for i, class_name in enumerate(classes): if class_name.lower() == "sajan": sajan_index = i break # ==================================================== # SAJAN CONFIDENCE # ==================================================== if sajan_index is not None: sajan_confidence = float( probabilities[0, sajan_index].item() ) else: sajan_confidence = 0.0 # ==================================================== # FINAL DECISION # ==================================================== if sajan_confidence >= SAJAN_THRESHOLD: prediction = "Sajan" else: prediction = "UNKNOWN" # ==================================================== # CONFIDENCE # ==================================================== confidence = round(sajan_confidence, 4) confidence_percent = round(sajan_confidence * 100, 2) # ==================================================== # SAVE RESULT DIRECTLY INSIDE MODEL ROUTE # ==================================================== file_exists = os.path.exists(CSV_FILE) with open(CSV_FILE, "a", newline="", encoding="utf-8") as f: writer = csv.writer(f) if not file_exists: writer.writerow([ "timestamp", "prediction", "confidence_percent", "face_detected" ]) writer.writerow([ datetime.now().strftime("%Y-%m-%d %H:%M:%S"), prediction, confidence_percent, True ]) # ==================================================== # RESPONSE # ==================================================== return jsonify({ "status": "success", "prediction": prediction, "confidence": confidence, "confidence_percent": confidence_percent, "face_detected": True, "bounding_box": { "x": int(x1), "y": int(y1), "width": int(x2 - x1), "height": int(y2 - y1) } }), 200 except Exception as e: return jsonify({ "status": "error", "message": str(e) }), 500 # ============================================================ # GET LAST STORED MODEL 0 RESULT # ============================================================ @app.route("/model/0/get", methods=["GET"]) def get_model_0_result(): CSV_FILE = "predictions.csv" if not os.path.exists(CSV_FILE): return jsonify({ "status": "success", "prediction": "UNKNOWN", "confidence_percent": 0, "face_detected": False, "source": "no_stored_result" }), 200 try: with open(CSV_FILE, "r", encoding="utf-8") as f: rows = list(csv.DictReader(f)) if not rows: return jsonify({ "status": "success", "prediction": "UNKNOWN", "confidence_percent": 0, "face_detected": False, "source": "empty" }), 200 last = rows[-1] return jsonify({ "status": "success", "timestamp": last["timestamp"], "prediction": last["prediction"], "confidence_percent": float(last["confidence_percent"]), "face_detected": last["face_detected"].lower() == "true", "source": "csv" }), 200 except Exception as e: return jsonify({ "status": "error", "message": str(e) }), 500 @app.route("/model/1", methods=["GET", "POST"]) def model_1(): if request.method == "GET": return jsonify({ "status": "model is running", "model": "Fruit and Vegetable Condition Classifier", "classes": ["healthy", "unhealthy"] }), 200 model_path = Path(__file__).parent / "models" / "fruit_veg.pt" image_size = (224, 224) class_labels = ("healthy", "unhealthy") max_image_pixels = 25_000_000 max_upload_bytes = 10 * 1024 * 1024 if request.content_length and request.content_length > max_upload_bytes: return jsonify({"error": "Image upload exceeds the 10 MB limit."}), 413 uploaded_image = request.files.get("image") if uploaded_image is None or not uploaded_image.filename: return jsonify({"error": "Send an image using the multipart field 'image'."}), 400 try: image_data = uploaded_image.stream.read(max_upload_bytes + 1) if len(image_data) > max_upload_bytes: return jsonify({"error": "Image upload exceeds the 10 MB limit."}), 413 with Image.open(BytesIO(image_data)) as image: image.verify() with Image.open(BytesIO(image_data)) as image: if image.width * image.height > max_image_pixels: return jsonify({"error": "Image dimensions are too large."}), 400 image = ImageOps.exif_transpose(image).convert("RGB") image = image.resize(image_size, Image.Resampling.BILINEAR) except (OSError, ValueError) as exc: return jsonify({ "status": "error", "message": f"Invalid image: {exc}" }), 400 if not model_path.is_file(): return jsonify({ "status": "error", "message": f"Model file not found: {model_path.name}" }), 503 try: tensor = torch.frombuffer( bytearray(image.tobytes()), dtype=torch.uint8 ) tensor = tensor.reshape(image_size[1], image_size[0], 3) tensor = tensor.permute(2, 0, 1).to(torch.float32).div_(255.0) mean = torch.tensor([0.485, 0.456, 0.406]).view(3, 1, 1) std = torch.tensor([0.229, 0.224, 0.225]).view(3, 1, 1) tensor = ((tensor - mean) / std).unsqueeze(0) model = torch.jit.load(str(model_path), map_location="cpu") model.eval() with torch.inference_mode(): logits = model(tensor) if logits.shape != (1, len(class_labels)): return jsonify({ "status": "error", "message": "Model returned an unexpected output." }), 500 probabilities = torch.softmax(logits, dim=1)[0] predicted_index = int(probabilities.argmax().item()) except (OSError, RuntimeError, ValueError) as exc: return jsonify({ "status": "error", "message": f"Prediction failed: {exc}" }), 500 prediction = class_labels[predicted_index] return jsonify( { "prediction": prediction, "confidence_percent": 100.0 } ), 200 # ============================================================ # START # ============================================================ if __name__ == "__main__": port = int(os.environ.get("PORT", 7860)) app.run( host="0.0.0.0", port=port, debug=False )