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 https://github.com/shiming-chen/GenZSL.

In the years since Goodfellow et al. introduced Generative Adversarial Networks (GANs) [1], there has been an explosion in the breadth and quality of generative model applications. Despite this work, GANs still have a long way to go before they see mainstream adoption, owing largely to their infamous training instability. Here I propose the Autoencoding Generative Adversarial Network (AEGAN), a four-network model which learns a bijective mapping between a specified latent space and a given sample space by applying an adversarial loss and a reconstruction loss to both the generated images and the generated latent vectors. The AEGAN technique offers several improvements to typical GAN training, including training stabilization, mode-collapse prevention, and permitting the direct interpolation between real samples. The effectiveness of the technique is illustrated using an anime face dataset.