Chapter 29: Object Detection

“Image classification tells you what—object detection tells you where.”

---

What Is Object Detection?

Object detection not only classifies what is in an image, but also where it is—by drawing bounding boxes around detected objects.

This enables applications like:

• Face detection in photos
• Pedestrian detection for self-driving cars
• Barcode readers in retail
• Real-time object tracking in surveillance systems

---

Classification vs Detection

Task	Input	Output
Image Classification	Image	Label (e.g., "dog")
Object Detection	Image	Label + Bounding Box

Example:

Classification: “There is a cat in the image.”

Detection: “There is a cat at coordinates (x1, y1) to (x2, y2).”

---

Tools for Object Detection in TensorFlow

Option 1: Pretrained Models via TensorFlow Hub

import tensorflow as tf
import tensorflow_hub as hub

detector = hub.load("https://tfhub.dev/google/fasterrcnn/openimages_v4/inception_resnet_v2/1")

Input an image:

import numpy as np
from PIL import Image

def load_image(path):
    img = Image.open(path).resize((512, 512))
    return np.array(img) / 255.0

image = load_image('sample.jpg')
result = detector(tf.convert_to_tensor([image]), training=False)

---

Option 2: TF Object Detection API (Advanced)

TensorFlow has a dedicated Object Detection API with models like:

• SSD (Single Shot MultiBox Detector)
• Faster R-CNN
• EfficientDet

⚠️ Requires setup with:

• Protobuf compilation
• COCO-trained checkpoints
• Custom label maps

---

Using a Pretrained SSD Model

For simpler use cases, try SSD from tf.keras.applications or prepackaged TF Hub models.

detector = hub.load("https://tfhub.dev/tensorflow/ssd_mobilenet_v2/2")

Visualize predictions:

import matplotlib.pyplot as plt
import tensorflow_datasets as tfds

image = tf.image.convert_image_dtype(tf.io.decode_jpeg(tf.io.read_file('cat.jpg')), tf.float32)
image_resized = tf.image.resize(image, (320, 320))

results = detector(tf.expand_dims(image_resized, axis=0))
print(results['detection_boxes'])

---

Visualizing Bounding Boxes

import matplotlib.patches as patches

def draw_boxes(image, boxes):
    fig, ax = plt.subplots(1)
    ax.imshow(image)
    for box in boxes:
        y1, x1, y2, x2 = box
        rect = patches.Rectangle((x1*image.shape[1], y1*image.shape[0]), 
                                 (x2 - x1)*image.shape[1], 
                                 (y2 - y1)*image.shape[0], 
                                 linewidth=2, edgecolor='red', facecolor='none')
        ax.add_patch(rect)
    plt.show()

---

Real-World Applications

Industry	Application
Automotive	Pedestrian/vehicle detection
Retail	Checkout-free shopping
Healthcare	Tumor boundary detection
Manufacturing	Faulty component identification

---

Try It Live

🔗 Colab Notebook: Object Detection with TF Hub
(Ready to be plugged into the companion repo)

---

Summary

In this chapter, you:

• Learned how object detection adds spatial awareness to classification
• Explored TensorFlow Hub and the Object Detection API
• Built a pipeline to detect objects and visualize bounding boxes

---