Open Internet by MindsNet

Improving Image Embeddings for Visual Appearance Similarity

The task is to develop image embeddings that focus on visual appearance, such as color and pattern, rather than object shape. Current methods, including VGG-based encoders, CLIP, and Color2Embed, have limitations in achieving this goal. A robust solution is needed to improve image similarity systems.

Computing & Technology, Computer Science, Machine Learning