In addition to the typical classification or metric learning loss, we also incorporate task-specific losses, such as predicting fashion attributes and color.
Building Lens your Look: Unifying text and camera search
Pinterest Engineering

Could you give more details on this part? I mean, do you (1) use a single loss function, train, use other loss function, train again with transfer learning, and repeat, or (2) use a ‘big’ loss function which comprises other functions for specific issues, such as fashion attributes, textures, colors, and so on?

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.