
Developed a background removal service for images, using advanced algorithms for logos and real-life images, achieving high accuracy.
A background removal service to remove background from images to be used for either embroidery or printing in apparel.
For logos with background, an algorithmic approach was used including a custom form of flood fill.
For real-life images, first object detection was used to separate the object. Then an ensemble of separate pre-existing segmentation models such as PointRend was used and on top of that GrabCut was applied to further improve the final output. In that case, an accuracy of about 97%-99% was achieved.