3D object detection in Computer Vision - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding 3D Bounding Boxes

In 3D object detection, what does a 3D bounding box typically represent?

A A box that encloses the object in 3D space with position, size, and orientation

B A 2D rectangle drawn around the object in an image

C A point cloud representing the object's surface

D A heatmap showing object presence probability

💡 Hint

Think about how to describe an object’s location and shape in three dimensions.

A 3D bounding box defines the object's position, size (width, height, length), and orientation in 3D space, enclosing it tightly.

❓ Model Choice

intermediate

Choosing a Model for 3D Object Detection from Point Clouds

Which model architecture is best suited for detecting objects directly from raw 3D point cloud data?

A Recurrent Neural Network (RNN) for sequential data

B PointNet, which processes unordered point sets directly

C Convolutional Neural Network (CNN) designed for 2D images

D Transformer model for natural language processing

💡 Hint

Consider which model can handle unordered 3D points without converting to images.

PointNet is designed to process unordered 3D point clouds directly, making it suitable for 3D object detection from raw point data.

❓ Predict Output

advanced

Output Shape of 3D Object Detection Network

Given a 3D object detection network that outputs a tensor of shape (batch_size, num_anchors, 7), what does the last dimension of size 7 represent?

A Class probabilities for 7 object categories

B Coordinates (x, y), width, height, depth, and confidence score

C RGB color channels and depth value

D Coordinates (x, y, z), dimensions (w, h, l), and rotation angle

💡 Hint

Think about what parameters define a 3D bounding box and its orientation.

The 7 values typically represent the 3D box center (x, y, z), size (width, height, length), and rotation angle around an axis.

❓ Metrics

advanced

Evaluating 3D Object Detection Performance

Which metric is commonly used to evaluate the accuracy of 3D object detection models?

A 3D Intersection over Union (IoU) between predicted and ground truth boxes

B Mean Squared Error (MSE) of pixel intensities

C BLEU score for language translation

D Confusion matrix for binary classification

💡 Hint

Consider how to measure overlap between predicted and true 3D boxes.

3D IoU measures the volume overlap between predicted and ground truth bounding boxes, indicating detection accuracy.

🔧 Debug

expert

Debugging a 3D Object Detection Model with Poor Orientation Predictions

A 3D object detection model predicts bounding boxes with accurate positions and sizes but consistently wrong rotation angles. What is the most likely cause?

A The model uses 2D convolution layers instead of 3D convolutions

B The input point cloud is missing color information

C The loss function does not properly penalize rotation angle errors

D The batch size is too large causing overfitting

💡 Hint

Think about what part of training controls orientation accuracy.

If rotation errors are high but other box parameters are good, the loss function likely does not weight rotation angle errors enough or uses an unsuitable representation.