Dataset bias in vision in Computer Vision - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is dataset bias in vision?

Dataset bias in vision happens when the images or data used to train a model do not represent the real world well. This causes the model to perform poorly on new or different images.

beginner

Why is dataset bias a problem for computer vision models?

Because biased datasets make models learn wrong or incomplete patterns. This leads to errors when the model sees new images that are different from the training data.

beginner

Name one common cause of dataset bias in vision datasets.

One common cause is collecting images from limited sources or environments, like only indoor photos or only one type of camera, which limits diversity.

intermediate

How can we reduce dataset bias in vision projects?

We can reduce bias by collecting diverse images from many sources, using data augmentation, and testing models on different datasets to check fairness.

intermediate

What is an example of dataset bias affecting a vision model in real life?

A face recognition system trained mostly on light-skinned faces may fail to recognize dark-skinned faces well, showing bias from the training data.

What does dataset bias in vision mainly affect?

A Size of the dataset

B Speed of model training

C Model's ability to generalize to new images

D Color of images

Dataset bias affects how well a model can apply what it learned to new, unseen images.

Which of these is a cause of dataset bias?

A Collecting images only from one camera type

B Increasing dataset size

C Adding noise to images

D Using images from many different environments

Using images from only one camera type limits diversity and can cause bias.

How can data augmentation help with dataset bias?

A By creating more diverse images from existing ones

B By removing images from the dataset

C By speeding up training

D By reducing image resolution

Data augmentation creates variations of images to increase diversity and reduce bias.

What is a sign that a vision model suffers from dataset bias?

A It performs well on all types of images

B It performs poorly on images different from training data

C It trains very fast

D It uses a lot of memory

Poor performance on new or different images indicates dataset bias.

Which approach helps test if a vision model is biased?

A Testing on the same dataset used for training

B Using fewer images

C Training longer

D Testing on a different, diverse dataset

Testing on a different dataset shows if the model generalizes well or is biased.

Explain what dataset bias in vision is and why it matters.

Think about how training data affects what the model learns.

Describe methods to identify and reduce dataset bias in vision datasets.

Consider both checking and fixing bias.