README.md

---
license: apache-2.0
tags:
- object-detection
- yolo
- yolov8
- onnx
- infrastructure
- wastemanagement
- flooding
---
# YOLOv8 Object Detection Model

This is a custom trained YOLOv8 nano model for detecting:
- **Flooding**
- **Potholes**
- **Waste Management**

The model was trained for 200 epochs on a custom dataset.

## Model Details
- **Architecture:** YOLOv8n (nano)
- **Input Size:** 640x640
- **Export Format:** ONNX
- **Training Dataset:** Custom dataset for 'Flooding', 'Potholes', 'Waste Management'
- **Metrics (best epoch 65 on validation set):**
  - mAP50-95: 0.393
  - mAP50: 0.606
  - Flooding mAP50: 0.797
  - Potholes mAP50: 0.430
  - Waste Management mAP50: 0.591

## Usage (Example)

This model is provided in ONNX format and can be used for inference in various environments. Here's how you might load it with ONNX Runtime in Python:

```python
import onnxruntime as ort
import numpy as np
from PIL import Image

# Load the ONNX model
session = ort.InferenceSession("Jin0908/muncipal-problem-detection/best.onnx") # Adjust path if downloaded locally

# Prepare input image (example: resize to 640x640, normalize)
image = Image.open("your_image.jpg").resize((640, 640))
input_array = np.array(image) / 255.0  # Normalize to [0, 1]
input_array = np.transpose(input_array, (2, 0, 1)) # (H, W, C) to (C, H, W)
input_array = np.expand_dims(input_array, axis=0).astype(np.float32) # Add batch dimension

# Run inference
outputs = session.run(None, {"images": input_array}) # Corrected: doubled curly braces for literal dict syntax

# Process outputs (this part is specific to YOLOv8 ONNX output and requires custom parsing)
# The output format is typically [batch_size, num_boxes, 4 + num_classes] or similar.
print("Model output shape:", outputs[0].shape)
# Further processing would involve non-max suppression (NMS) and decoding bounding boxes.
```