The Effects of color spaces in Deep Learning
One of the largest applications of Deep Learning is Computer Vision, particularly Image Classification, which consists in analyze digital images, that are represented in a particular color space. Color spaces attempt to simulate the spectrum of the human color vision, representing the images in terms of saturation, illumination, and hue values. Recent experiments suggest that color spaces can significantly affect the classification accuracy of deep convolutional networks.
The standard color space for digital images is RBG, but it is not the only one. Depending on the application, an image might be better represented in a different color space, therefore experimenting with color spaces may help you get better results.
The outline for this talk is as follows:
- Introduction (1 min)
- Theory of color space (5 min)
- Standard color spaces (3 min)
- Color spaces in Python (5 min)
- Choice for color spaces for different applications in real-world (3 min)
- Overview of recent experiments and open questions (3 min)
- Takeaways (2 min)
- Q/A (5 min)
This talk will be more valuable to novel and intermediate AI practitioners, who have a general understanding of deep learning and computer vision.
Sara is a seasoned software developer and data enthusiast based in Guatemala. She loves Python and is actively involved with the tech community. Sara is particularly interested in advocating for women's inclusion and supporting young women in pursuing careers in STEM.