Learning Conditional Attributes for Compositional Zero-Shot Learning
This work addresses a key problem in AI for recognizing novel compositional concepts, offering an incremental improvement in modeling attribute-object interactions for zero-shot learning tasks.
The paper tackles the challenge in Compositional Zero-Shot Learning of modeling attributes that interact differently with various objects, such as 'wet' in 'wet apple' versus 'wet cat', by proposing a framework to learn conditional attribute embeddings. Experiments on benchmarks, including the challenging C-GQA dataset, show improved performance over state-of-the-art methods, validating the approach's effectiveness.
Compositional Zero-Shot Learning (CZSL) aims to train models to recognize novel compositional concepts based on learned concepts such as attribute-object combinations. One of the challenges is to model attributes interacted with different objects, e.g., the attribute ``wet" in ``wet apple" and ``wet cat" is different. As a solution, we provide analysis and argue that attributes are conditioned on the recognized object and input image and explore learning conditional attribute embeddings by a proposed attribute learning framework containing an attribute hyper learner and an attribute base learner. By encoding conditional attributes, our model enables to generate flexible attribute embeddings for generalization from seen to unseen compositions. Experiments on CZSL benchmarks, including the more challenging C-GQA dataset, demonstrate better performances compared with other state-of-the-art approaches and validate the importance of learning conditional attributes. Code is available at https://github.com/wqshmzh/CANet-CZSL