
AI: Advancing one pixel at a time
The things we do for Love ;) Valar Morghulis. For many AI is just a buzz word or USP but for us, it’s more than the funky marketing stuff. Believe me when I say: we dream in pixels!
Parsing out the fine information (in this case clothing style) from the crowded/wild background images is a daunting task for any state of the art deep learning tools. Chalking out detailed semantic relationship within pixels is one of the problem on which some of the top deep learning labs are working on. The difficulty with such images is that they have wild background, are ill posed and have variance in illumination along the surface of clothing. To segment such scattered information in a detailed way is a really difficult task. But happy to share that we have made significant advances towards solving this problem and are planning to make this research (wrapped around some product) live soon.
The dark purple area in the image (right) shows the segmented clothing region through our developed neural network which can further be used to extract the styling (other than color and pattern) of apparels in real time in a more robust way.
P.S. Science advances in small steps and we should stay humble!