Assignment 7: Computer Vision Research
Name three applications of Computer Vision you can recognize on your phone
Face unlock (biometric security): Uses facial recognition to securely unlock your device.
Augmented reality filters on apps like Snapchat or Instagram: Real-time face detection and overlay effects.
Image recognition in photo galleries: Grouping pictures by people or objects automatically.
Take a photo of your desk and ask ChatGPT to identify or name all the objects
ChatGPT identified everything correctly! It found my laptop, my bottle of water, and 2 books.
How many steps are involved in this identification process?
About 3 steps:
Image preprocessing: Making the picture workable for the model.
Feature extraction: Detecting shapes, colors, and patterns.
Classification/labeling: Matching features to the objects that they belong to.
Can you identify any challenges for ChatGPT?
In the context of computer vision, ChatGPT has many challenges: it often cannot adjust to unequal lighting, ambiguous objects (e.g. book vs notebook that look similar), and items it had rarely seen before (because ChatGPT heavily relies on its training data, so if it underrepresented a certain object, it would struggle to classify it).
Ask ChatGPT to highlight or circle a specific item on your desk. Describe the results and possible challenges
I asked it to circle a bottle of water on my desk, and it drew a very big circle that also included a book on my desk. Hence, it was correct but suboptimal, which represents a big challenge in object detection - even if you circle the object correctly, it may not be the best/smallest circle possible.
Several applications involve decision-making processes that rely on Computer Vision. Name some applications that you believe are not trustworthy or safe.
Facial recognition for law enforcement: High risk of bias (e.g. racial and gender bias) and misidentification.
Autonomous weapons: Decisions of life and death shouldn't be outsourced to vision systems.
Emotion detection from faces: Again, high risk of bias and misclassification, as every person's emotion is very different.