GenZSL: Generative Zero-Shot Learning Via Inductive Variational Autoencoder

Remarkable progress in zero-shot learning (ZSL) has been achieved using generative models. However, existing generative ZSL methods merely generate (imagine) the visual features from scratch guided by the strong class semantic vectors annotated by experts, resulting in suboptimal generative performance and limited scene generalization. To address these and advance ZSL, we propose an inductive variational autoencoder for generative zero-shot learning, dubbed GenZSL. Mimicking human-level concept learning, GenZSL operates by inducting new class samples from similar seen classes using weak class semantic vectors derived from target class names (i.e., CLIP text embedding). To ensure the generation of informative samples for training an effective ZSL classifier, our GenZSL incorporates two key strategies. Firstly, it employs class diversity promotion to enhance the diversity of class semantic vectors. Secondly, it utilizes target class-guided information boosting criteria to optimize the model. Extensive experiments conducted on three popular benchmark datasets showcase the superiority and potential of our GenZSL with significant efficacy and efficiency over f-VAEGAN, e.g., 24.7\% performance gains and more than $60\times$ faster training speed on AWA2. Codes are available at \url{https://github.com/shiming-chen/GenZSL}.

We propose an inductive variational autoencoder as a new generative model for zero-shot learning, namely GenZSL. Inspired by human perception, GenZSL operates on an induction-based approach to synthesize informative and high-quality sample features for unseen classes. To achieve this, we introduce class diversity promotion to enhance the diversity and discrimination of class semantic vectors. Additionally, we design two losses targeting the criteria of target class-guided information boosting to optimize the model.Through qualitative and quantitative analyses, we demonstrate that GenZSL consistently outperforms existing generative ZSL methods in terms of efficacy and efficiency.