Obtain the model predictions on the test dataset. Create a list of computed IoUs for each class. Each list should have a length equal to the size of the test dataset. If there is no predicted bounding box on a particular image, put None instead of the IoU.

# Import compute$\_$iou function from Keras CV

from keras$\_$cv$.$bounding$\_$box import compute$\_$iou

# Get the model predictions. This will take some time.

# Save the model predictions to objdet$\_$preds variable

### $--$YOUR CODE HERE$--$ ###

## I KEEP GETTING ERRORS HERE

# Extract the predicted and ground truth boxes for each object class

### $--$YOUR CODE HERE$--$ ###

class$\_$banana$\_$gt $=$ ds$\_$test$\_$sample$\_$batch$[0,...].$numpy$()$

#class$\_$apple$\_$gt $=$ ds$\_$test$\_$sample$\_$images$["$bounding$\_$boxes"$][1]$

#class$\_$orange$\_$gt $=$ ds$\_$test$\_$sample$\_$images$["$bounding$\_$boxes"$][2]$

# Compute the IoU of the bounding boxes for each object in each image

# Set the appropriate boxes parameters

# There should be three IoU lists

# Iterate through the prediction output of the model

# Save them to apple$\_$ious, banana$\_$ious, and orange$\_$ious variables, respectively

### $--$YOUR CODE HERE$--$ ###

apple$\_$ious $=[]$

banana$\_$ious $=[]$

orange$\_$ious $=[]$

for i in range$($len$($objdet$\_$preds$))$:

pred$\_$boxes $=$ objdet$\_$preds$[$i$]["$boxes$"]$

gt$\_$boxes$\_$apple $=$ class$\_$apple$\_$gt$[$i$]$

gt$\_$boxes$\_$banana $=$ class$\_$banana$\_$gt$[$i$]$

gt$\_$boxes$\_$orange $=$ class$\_$orange$\_$gt$[$i$]$

apple$\_$iou $=[]$

banana$\_$iou $=[]$

orange$\_$iou $=[]$

for j in range$($len$($pred$\_$boxes$))$:

apple$\_$iou.append$($compute$\_$iou$($pred$\_$boxes$[$j$],$ gt$\_$boxes$\_$apple$))$

banana$\_$iou.append$($compute$\_$iou$($pred$\_$boxes$[$j$],$ gt$\_$boxes$\_$banana$))$

orange$\_$iou.append$($compute$\_$iou$($pred$\_$boxes$[$j$],$ gt$\_$boxes$\_$orange$))$

apple$\_$ious.append$($apple$\_$iou$)$

banana$\_$ious.append$($banana$\_$iou$)$

orange$\_$ious.append$($orange$\_$iou$)$