SSD concept in Computer Vision - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to create a default box (anchor) size for SSD.

Computer Vision

default_box_size = [1]

Drag options to blanks, or click blank then click option'

A2.0

B0.2

C1.0

D0.5

Attempts:

3 left

💡 Hint

Common Mistakes

Using a size larger than 1.0 which is bigger than the image.

Choosing 0.5 which is quite large for a default box.

The default box size is usually a fraction of the image size, such as 0.2 (20%).

2fill in blank

medium

Complete the code to calculate the number of default boxes per feature map location in SSD.

Computer Vision

num_default_boxes = [1]

Drag options to blanks, or click blank then click option'

A8

B4

C6

D10

Attempts:

3 left

💡 Hint

Common Mistakes

Choosing 4 which is too few to cover aspect ratios.

Choosing 10 which is more than typical SSD.

SSD typically uses 6 default boxes per location to cover different aspect ratios.

3fill in blank

hard

Fix the error in the code to compute the scale of the default box for the k-th feature map layer in SSD.

Computer Vision

scale_k = s_min + (s_max - s_min) * [1] / (m - 1)

Drag options to blanks, or click blank then click option'

Ak

Bk + 1

Ck - 1

Dm

Attempts:

3 left

💡 Hint

Common Mistakes

Using 'k + 1' which makes the first scale larger than s_min.

Using 'k - 1' which makes early scales too small or negative.

Using 'm' which does not vary with k.

The standard formula assumes k starts from 0: s_k = s_min + (s_max - s_min) * k / (m - 1), giving s_min for first layer and s_max for last.

4fill in blank

hard

Fill both blanks to create a dictionary comprehension that maps each default box to its aspect ratio and scale in SSD.

Computer Vision

default_boxes = {i: ([1], [2]) for i in range(num_default_boxes)}

Drag options to blanks, or click blank then click option'

Aaspect_ratios[i]

Bscales[i]

Caspect_ratios

Dscales

Attempts:

3 left

💡 Hint

Common Mistakes

Using the whole list instead of indexed element.

Swapping aspect_ratios and scales.

Each default box uses the i-th aspect ratio and scale from their lists.

5fill in blank

hard

Fill all three blanks to filter predicted boxes with confidence above threshold and sort by confidence in SSD post-processing.

Computer Vision

filtered_boxes = [(box, conf) for box, conf in zip(predicted_boxes, confidences) if conf [1] [2]]
sorted_boxes = sorted(filtered_boxes, key=lambda x: x[[3]], reverse=True)

Drag options to blanks, or click blank then click option'

A>

B0.5

C1

D<

Attempts:

3 left

💡 Hint

Common Mistakes

Using '<' instead of '>' for filtering.

Using wrong index for sorting key.

We keep (box, conf) pairs with confidence > 0.5 and sort by confidence at index 1.