feat: deploy PPE detection backend to HF Spaces (YOLO11s ONNX)

.gitignore

@@ -0,0 +1,2 @@
+__pycache__/
+*.pyc

README.md

@@ -1,14 +1,49 @@
 ---
-title: PPE
-emoji: 🔥
+title: PPE Compliance Detector
+emoji: 🦺
 colorFrom: yellow
-colorTo: purple
+colorTo: orange
 sdk: gradio
-sdk_version: 6.10.0
+sdk_version: 5.12.0
 app_file: app.py
 pinned: false
-license: apache-2.0
-short_description: PPE detection model
+license: mit
+short_description: Detect missing hard hats and safety vests on construction sites
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# 🦺 PPE Compliance Detector
+
+Detect whether construction site workers are wearing required Personal Protective Equipment
+(hard hats and high-visibility vests) from images.
+
+## Model
+
+- **Architecture:** YOLO11s (Ultralytics)
+- **Format:** ONNX (CPU-optimised)
+- **Training:** 100 epochs, imgsz=1280, RTX 4060
+- **Test mAP50:** 93.2%
+- **Minority class AP50:** >91% (vest, no-vest)
+
+## Classes
+
+| Class | Description |
+|---|---|
+| `hardhat` | Worker wearing a hard hat |
+| `no-hardhat` | Worker without a hard hat ⚠️ |
+| `vest` | Worker wearing a high-vis vest |
+| `no-vest` | Worker without a high-vis vest ⚠️ |
+| `person` | Full body of a worker |
+
+## Training Data
+
+Merged from three public datasets (~10K images):
+- Construction Site Safety (Roboflow)
+- SHWD — Safety Helmet Wearing Dataset (GitHub)
+- Pictor-PPE (GitHub)
+
+## Usage
+
+Upload a construction site image and the model will:
+1. Detect all workers and PPE items
+2. Draw bounding boxes with class labels
+3. Generate a compliance summary highlighting violations

app.py

@@ -0,0 +1,116 @@
+import gradio as gr
+from ultralytics import YOLO
+from PIL import Image
+import numpy as np
+import json
+import os
+
+# ---- Load model ----
+MODEL_PATH = os.path.join(os.path.dirname(__file__), "best.onnx")
+model = YOLO(MODEL_PATH)
+
+CLASS_NAMES = ["hardhat", "no-hardhat", "vest", "no-vest", "person"]
+VIOLATION_CLASSES = {"no-hardhat", "no-vest"}
+COMPLIANT_CLASSES = {"hardhat", "vest"}
+
+
+def detect_ppe(image, conf_threshold=0.25):
+    """Run PPE detection on an uploaded image."""
+    if image is None:
+        return None, "No image provided."
+
+    # Run inference
+    results = model(image, imgsz=640, conf=conf_threshold, verbose=False)
+    result = results[0]
+
+    # Draw annotated image
+    annotated = result.plot()
+    annotated_rgb = Image.fromarray(annotated[..., ::-1])
+
+    # Build compliance summary
+    detections = result.boxes
+    violations = []
+    compliant = []
+    persons = 0
+
+    for box in detections:
+        cls_name = CLASS_NAMES[int(box.cls)]
+        conf = float(box.conf)
+
+        if cls_name in VIOLATION_CLASSES:
+            violations.append(f"{cls_name} ({conf:.0%})")
+        elif cls_name in COMPLIANT_CLASSES:
+            compliant.append(f"{cls_name} ({conf:.0%})")
+        elif cls_name == "person":
+            persons += 1
+
+    summary = ""
+    if violations:
+        summary += f"\u26a0\ufe0f VIOLATIONS DETECTED ({len(violations)}):\n"
+        for v in violations:
+            summary += f"  \u274c {v}\n"
+        summary += "\n"
+    else:
+        summary += "\u2705 No PPE violations detected.\n\n"
+
+    if compliant:
+        summary += f"PPE Compliant Items ({len(compliant)}):\n"
+        for c in compliant:
+            summary += f"  \u2705 {c}\n"
+        summary += "\n"
+
+    summary += f"Workers detected: {persons}\n"
+    summary += f"Total detections: {len(detections)}"
+
+    # Build JSON result for API consumers (Vercel frontend)
+    api_result = {
+        "violations": violations,
+        "compliant": compliant,
+        "persons": persons,
+        "total_detections": len(detections),
+        "boxes": [],
+    }
+    for box in detections:
+        api_result["boxes"].append({
+            "class": CLASS_NAMES[int(box.cls)],
+            "confidence": round(float(box.conf), 4),
+            "bbox": box.xyxy[0].tolist(),
+        })
+
+    return annotated_rgb, summary
+
+
+# ---- Build Gradio interface ----
+example_dir = os.path.join(os.path.dirname(__file__), "examples")
+examples = []
+if os.path.isdir(example_dir):
+    for f in sorted(os.listdir(example_dir)):
+        if f.lower().endswith((".jpg", ".jpeg", ".png")):
+            examples.append([os.path.join(example_dir, f)])
+
+demo = gr.Interface(
+    fn=detect_ppe,
+    inputs=[
+        gr.Image(type="numpy", label="Upload Construction Site Image"),
+        gr.Slider(
+            minimum=0.1, maximum=0.9, value=0.25, step=0.05,
+            label="Confidence Threshold",
+        ),
+    ],
+    outputs=[
+        gr.Image(label="Detection Result"),
+        gr.Textbox(label="Compliance Summary", lines=10),
+    ],
+    title="\U0001f9ba PPE Compliance Detector",
+    description=(
+        "Upload a construction site image to detect hard hats and safety vests. "
+        "The model identifies PPE violations (missing hard hat or vest) and "
+        "highlights them with bounding boxes.\n\n"
+        "**Model:** YOLO11s (ONNX) \u2014 93.2% test mAP50 | "
+        "**Classes:** hardhat, no-hardhat, vest, no-vest, person"
+    ),
+    examples=examples if examples else None,
+    cache_examples=False,
+)
+
+demo.launch()

best.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3d600afc077aac988a25f6f51a04f80a81857fa30943fc91fb11eebc0afe1e4
+size 37934347

model_config.json

@@ -0,0 +1,14 @@
+{
+  "class_names": ["hardhat", "no-hardhat", "vest", "no-vest", "person"],
+  "num_classes": 5,
+  "violation_classes": ["no-hardhat", "no-vest"],
+  "compliant_classes": ["hardhat", "vest"],
+  "model": {
+    "architecture": "YOLO11s",
+    "format": "ONNX",
+    "input_size": 640,
+    "training_imgsz": 1280,
+    "test_map50": 0.932,
+    "test_map5095": 0.639
+  }
+}

requirements.txt

@@ -0,0 +1,7 @@
+# PPE Compliance Detector — HuggingFace Spaces
+# Runtime: Python 3.11, CPU only
+ultralytics>=8.3.50
+onnxruntime>=1.20.1
+opencv-python-headless>=4.10.0
+Pillow>=11.1.0
+numpy<2.0