CIZSL++: Creativity Inspired Generative Zero-Shot Learning
This work addresses the problem of recognizing unseen categories in zero-shot learning, which is important for applications where new classes frequently emerge without labeled data.
This paper introduces CIZSL++, a creativity-inspired generative zero-shot learning model. It models the visual learning process of unseen categories by drawing inspiration from human creativity, aiming to improve the discriminative power of ZSL. The CIZSL losses consistently improve generative ZSL models on generalized ZSL from noisy text on CUB and NABirds datasets, and also show advantages for attribute-based ZSL on AwA2, aPY, and SUN datasets.
Zero-shot learning (ZSL) aims at understanding unseen categories with no training examples from class-level descriptions. To improve the discriminative power of ZSL, we model the visual learning process of unseen categories with inspiration from the psychology of human creativity for producing novel art. First, we propose CIZSL-v1 as a creativity inspired model for generative ZSL. We relate ZSL to human creativity by observing that ZSL is about recognizing the unseen, and creativity is about creating a likable unseen. We introduce a learning signal inspired by creativity literature that explores the unseen space with hallucinated class-descriptions and encourages careful deviation of their visual feature generations from seen classes while allowing knowledge transfer from seen to unseen classes. Second, CIZSL-v2 is proposed as an improved version of CIZSL-v1 for generative zero-shot learning. CIZSL-v2 consists of an investigation of additional inductive losses for unseen classes along with a semantic guided discriminator. Empirically, we show consistently that CIZSL losses can improve generative ZSL models on the challenging task of generalized ZSL from a noisy text on CUB and NABirds datasets. We also show the advantage of our approach to Attribute-based ZSL on AwA2, aPY, and SUN datasets. We also show that CIZSL-v2 has improved performance compared to CIZSL-v1.