Object Detection $($I WANT CODING$)$

$1.$ Dataset: Use the oxford$\_$iiit$\_$pet dataset from Keras or PyTorch.

$$ For object detection, you will need to find bounding boxes that correspond to the pet

objects in the segmentation masks. To create a bounding box for each pet, you can:

$$ Identify the non$-$background pixels in the segmentation mask.

$$ Compute the minimum and maximum coordinates $($xmin$,$ xmax, ymin, ymax$)$ that

enclose all the pet pixels. This will give you a bounding box for each image around the

pet object.

$$ Example of how to calculate bounding boxes from segmentation masks:

import numpy as np

def get$\_$bounding$\_$boxsegmentation$\_$mask$)$:

# Find non$-$zero regions in the segmentation mask $($ i $.$ e $.,$ the pet area$)$

pet$\_$pixels $=$ np$.$ argwhere$($segmentation$\_$mask $0)$

# Compute the bounding box coordinates

ymin, xmin $=$ np$.$min$($pet$\_$pixels, axis$=0)$

ymax, xmax $=.$max$($pet$\_$pixels, axis$=0)$

r e t u r n xmin, ymin, xmax, max

$2.$ Using a Pre$-$trained YOLO Model:

$$ Download a pre$-$trained YOLO model: YOLOv$3$ Model

$$ Convert the model weights weights to a Keras$-$compatible h$5$ file.

$$ Load the model and test it on an image by normalizing the image and applying the model

for predictions.

$$ Show the detection results using bounding boxes drawn on the image.

$3.$ Starter Code for Pre$-$trained YOLOv$3$ Model Loading:

from tensorflow.keras.models import load$\_$model

import numpy as np

import cv$2$

import matplotlib.pyplot as plt

# Load the pre$-$trained YOLOv$3$ model

model $=$ load$\_$model $(\text{'}$yolov$3.$h$5\text{'})$

# Load and preprocess an image

def load$\_$image$($img$\_$path, target$\_$size$)$:

image $=$ cv$2.$ imread$($img$\_$path$)$

image $=$ cv$2.$ cvtColor $($image$,$ cv$2.$ COLOR$\_$BGR$2$RGB$)$

image$\_$resized $=$ cv$2.$ resize$($image$,$ target$\_$size$)$

image$\_$resized $=$ image$\_$resized $/255\mathrm{.}0$ # Normalize to $[0,1]$

return np$.$expand$\_$dims$($image$\_$resized, axis$=0),$ image

# Example image path and s i z e

img$\_$path $=$ 'path$\_$to$\_$your$\_$image.jpg$\text{'}$

input$\_$size $=(416,416)$ # Standard size for YOLOv$3$

# Load and preprocess image

input$\_$image, original$\_$image $=$ load$\_$image$($img$\_$path, input$\_$size$)$

# Make predictions using YOLO

predictions $=$ model. predict$($input$\_$image$)$

# Post$-$process and v i s u a l i z e the detection r e s u l t s

def draw$\_$boxes$($image$,$ boxes$)$:

for box in boxes:

r e t u r n xmin, ymin, xmax, max $=$ box

cv$2.$ rectangle$($image$,($xmin$,$ ymin$),($xmax$,$ ymax$),(0,255,0),2)$

image

# Example bounding boxes $($dummy values for i l l u s t r a t i o n $)$

boxes $=[[100,150,300,350]]$ # Replace with actual post$-$processed YOLO output

output$\_$image $=$ draw$\_$boxes$($original$\_$image.copy$),$ boxes$)$

# Display the output

plt$.$ imshow$($output$\_$image$)$

p l t $.$ a x i s $(\text{'}$ o ff $\text{'})$

plt$.$ show$()$

$4.$ Evaluation:

$$ Evaluate the model's performance on the oxford$\_$iiit$\_$pet dataset using the known

bounding boxes

$$ Calculate mAP at different loU thresholds $($mAP$\{0\mathrm{.}25\},$ mAP$\{0\mathrm{.}5\},$ mAP$\{0\mathrm{.}75\},$ m$\{0\mathrm{.}95\}).$