Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,120 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import cv2
|
| 3 |
+
import math
|
| 4 |
+
import torch
|
| 5 |
+
import numpy as np
|
| 6 |
+
from flask import Flask, request, jsonify, render_template_string
|
| 7 |
+
from ultralytics import YOLO
|
| 8 |
+
|
| 9 |
+
# ------------------------------
|
| 10 |
+
# CONFIG
|
| 11 |
+
# ------------------------------
|
| 12 |
+
MODEL_PATH = "best.pt" # your 5MB lightweight model
|
| 13 |
+
FOCAL_LENGTH_PX = 615
|
| 14 |
+
KNOWN_WIDTHS_M = {
|
| 15 |
+
"person": 0.5, "car": 1.8, "truck": 2.3, "bus": 2.5,
|
| 16 |
+
"bicycle": 0.6, "motorcycle": 0.7, "door": 0.9,
|
| 17 |
+
"stairs": 1.2, "pole": 0.15, "bollard": 0.2
|
| 18 |
+
}
|
| 19 |
+
THRESHOLDS = {"CRITICAL": 1.0, "WARNING": 2.0, "CAUTION": 3.0}
|
| 20 |
+
|
| 21 |
+
# ------------------------------
|
| 22 |
+
# Initialize Flask & YOLO
|
| 23 |
+
# ------------------------------
|
| 24 |
+
app = Flask(__name__)
|
| 25 |
+
device = 0 if torch.cuda.is_available() else "cpu"
|
| 26 |
+
model = YOLO(MODEL_PATH)
|
| 27 |
+
model.to(device)
|
| 28 |
+
|
| 29 |
+
# ------------------------------
|
| 30 |
+
# Utility Functions
|
| 31 |
+
# ------------------------------
|
| 32 |
+
def estimate_distance(bbox_width_px, class_name):
|
| 33 |
+
if bbox_width_px <= 1:
|
| 34 |
+
return None
|
| 35 |
+
known_width = KNOWN_WIDTHS_M.get(class_name)
|
| 36 |
+
if not known_width:
|
| 37 |
+
return None
|
| 38 |
+
return (known_width * FOCAL_LENGTH_PX) / float(bbox_width_px)
|
| 39 |
+
|
| 40 |
+
def get_alert_level(distance_m):
|
| 41 |
+
if distance_m is None:
|
| 42 |
+
return "SAFE"
|
| 43 |
+
if distance_m <= THRESHOLDS["CRITICAL"]:
|
| 44 |
+
return "CRITICAL"
|
| 45 |
+
if distance_m <= THRESHOLDS["WARNING"]:
|
| 46 |
+
return "WARNING"
|
| 47 |
+
if distance_m <= THRESHOLDS["CAUTION"]:
|
| 48 |
+
return "CAUTION"
|
| 49 |
+
return "SAFE"
|
| 50 |
+
|
| 51 |
+
# ------------------------------
|
| 52 |
+
# API ENDPOINT: /detect
|
| 53 |
+
# ------------------------------
|
| 54 |
+
@app.route("/detect", methods=["POST"])
|
| 55 |
+
def detect_objects():
|
| 56 |
+
"""
|
| 57 |
+
Accepts an image file (from Unity or mobile camera)
|
| 58 |
+
and returns detections + alert level JSON.
|
| 59 |
+
"""
|
| 60 |
+
if "file" not in request.files:
|
| 61 |
+
return jsonify({"error": "No file uploaded"}), 400
|
| 62 |
+
|
| 63 |
+
file = request.files["file"]
|
| 64 |
+
img_bytes = np.frombuffer(file.read(), np.uint8)
|
| 65 |
+
frame = cv2.imdecode(img_bytes, cv2.IMREAD_COLOR)
|
| 66 |
+
|
| 67 |
+
results = model(frame, conf=0.25, iou=0.5, verbose=False)
|
| 68 |
+
boxes = results[0].boxes
|
| 69 |
+
|
| 70 |
+
detections = []
|
| 71 |
+
max_level = "SAFE"
|
| 72 |
+
order = {"SAFE": 0, "CAUTION": 1, "WARNING": 2, "CRITICAL": 3}
|
| 73 |
+
|
| 74 |
+
for b in boxes:
|
| 75 |
+
x1, y1, x2, y2 = b.xyxy[0].tolist()
|
| 76 |
+
cls_id = int(b.cls[0].item())
|
| 77 |
+
conf = float(b.conf[0].item())
|
| 78 |
+
class_name = model.names.get(cls_id, str(cls_id))
|
| 79 |
+
bbox_w = x2 - x1
|
| 80 |
+
distance_m = estimate_distance(bbox_w, class_name)
|
| 81 |
+
level = get_alert_level(distance_m)
|
| 82 |
+
|
| 83 |
+
if order[level] > order[max_level]:
|
| 84 |
+
max_level = level
|
| 85 |
+
|
| 86 |
+
detections.append({
|
| 87 |
+
"class": class_name,
|
| 88 |
+
"confidence": round(conf, 3),
|
| 89 |
+
"distance_m": round(distance_m, 2) if distance_m else None,
|
| 90 |
+
"alert_level": level,
|
| 91 |
+
"bbox": [int(x1), int(y1), int(x2), int(y2)]
|
| 92 |
+
})
|
| 93 |
+
|
| 94 |
+
return jsonify({
|
| 95 |
+
"alert_level": max_level,
|
| 96 |
+
"detections": detections
|
| 97 |
+
})
|
| 98 |
+
|
| 99 |
+
# ------------------------------
|
| 100 |
+
# SIMPLE FRONTEND PAGE
|
| 101 |
+
# ------------------------------
|
| 102 |
+
@app.route("/")
|
| 103 |
+
def index():
|
| 104 |
+
return render_template_string("""
|
| 105 |
+
<html>
|
| 106 |
+
<head><title>YOLOv8 Real-Time Alerts</title></head>
|
| 107 |
+
<body style="font-family:sans-serif;text-align:center;">
|
| 108 |
+
<h2>YOLOv8 Object Detection API</h2>
|
| 109 |
+
<p>Send a POST /detect request with an image to get JSON alerts.</p>
|
| 110 |
+
<p>Example (curl):</p>
|
| 111 |
+
<code>
|
| 112 |
+
curl -X POST -F "file=@image.jpg" https://YOUR_SPACE_URL.hf.space/detect
|
| 113 |
+
</code>
|
| 114 |
+
</body>
|
| 115 |
+
</html>
|
| 116 |
+
""")
|
| 117 |
+
|
| 118 |
+
if __name__ == "__main__":
|
| 119 |
+
port = int(os.environ.get("PORT", 7860))
|
| 120 |
+
app.run(host="0.0.0.0", port=port)
|