Context-Aware Zero-Shot Recognition
This work addresses the problem of recognizing unseen objects in images for computer vision applications, offering an incremental improvement by incorporating context into existing zero-shot learning techniques.
The paper tackles zero-shot object recognition and detection by leveraging visual context and geometric relationships between objects, integrating this into a Conditional Random Field framework. Results on the Visual Genome dataset show significant performance improvements compared to traditional methods.
We present a novel problem setting in zero-shot learning, zero-shot object recognition and detection in the context. Contrary to the traditional zero-shot learning methods, which simply infers unseen categories by transferring knowledge from the objects belonging to semantically similar seen categories, we aim to understand the identity of the novel objects in an image surrounded by the known objects using the inter-object relation prior. Specifically, we leverage the visual context and the geometric relationships between all pairs of objects in a single image, and capture the information useful to infer unseen categories. We integrate our context-aware zero-shot learning framework into the traditional zero-shot learning techniques seamlessly using a Conditional Random Field (CRF). The proposed algorithm is evaluated on both zero-shot region classification and zero-shot detection tasks. The results on Visual Genome (VG) dataset show that our model significantly boosts performance with the additional visual context compared to traditional methods.