Field-Guide-Inspired Zero-Shot Learning
This reduces annotation costs for expert domains, making zero-shot learning more practical for real-world deployment, though it is incremental as it builds on existing attribute-based methods.
The paper tackles the problem of expensive expert annotation for zero-shot learning by introducing an interactive approach that asks for the most useful attributes, achieving performance comparable to full annotations with significantly fewer annotations on benchmarks like CUB, SUN, and AWA2.
Modern recognition systems require large amounts of supervision to achieve accuracy. Adapting to new domains requires significant data from experts, which is onerous and can become too expensive. Zero-shot learning requires an annotated set of attributes for a novel category. Annotating the full set of attributes for a novel category proves to be a tedious and expensive task in deployment. This is especially the case when the recognition domain is an expert domain. We introduce a new field-guide-inspired approach to zero-shot annotation where the learner model interactively asks for the most useful attributes that define a class. We evaluate our method on classification benchmarks with attribute annotations like CUB, SUN, and AWA2 and show that our model achieves the performance of a model with full annotations at the cost of a significantly fewer number of annotations. Since the time of experts is precious, decreasing annotation cost can be very valuable for real-world deployment.