Zero-Shot learning
Zero-Shot learning strong method to solve a task without receiving any example of that task at training phase.
Zero-shot learning aims to minimize the annotation requirements by enabling recognition of unseen classes, i.e. those with no training examples. This is achieved by transferring knowledge from seen to unseen classes by means of auxiliary data, typically obtained easily from textual sources.
Results on AWA
| Ours | Lampert group | |
|---|---|---|
| AWA1 | 60% |
44.1% |