Purpose: To combine object detection and image classification models to build a complete AI processing pipeline that locates, crops, and analyzes specific visual data in real time.

Mini Project: Automated Object Inspection Pipeline

---

• Students construct a multi-model pipeline that detects an object in real time, isolates its bounding region by cropping it, and displays both the visual crop and the detection output on screen.

Exercise 1 - Step 1: Recall the object detection model.

---

• Initialize the pre-trained object detection model (e.g., COCO-SSD) on the live webcam feed using ml5.js and p5.js.
• Draw bounding boxes and label tags around detected objects using their bounding box coordinates $(x, y, w, h)$.

Exercise 1 - Step 2: Crop the detected object from the output image and display the cropped image along with the detection result.

---

• Extract the pixel area corresponding to the primary detected bounding box using p5.js image extraction (`get(x, y, w, h)`).
• Render the dynamically cropped object snippet on screen in a dedicated preview box alongside the main webcam view and confidence scores.

Exercise 1 - Step 3: Pass the cropped image to a custom classifier.

---

• Pass the live cropped image directly as input into a second pre-trained or custom image classifier model.
• Display the final pipeline classification result below the cropped image.

Solutions

---

• Link to view the solution for exercise-1
• Link to view the solution code for exercise-1

1. Introduce Model Pipelining Concepts to students - passing output from one model as input to another model, and how this can be used to build more complex AI systems.
2. Explain latency and processing sequence: Model 1 must finish predicting before Model 2 can receive its input.
3. Points to Ponder:
   • What happens if the first model fails to detect an object? How can we handle this gracefully in the pipeline?