External knowledge (a.k.a side information) plays a critical role in zero-shot learning (ZSL) which aims to predict with unseen classes that have never appeared in training data. Several kinds of external knowledge such as text and attribute have been widely investigated, but they alone are limited with incomplete semantics. Therefore, some very recent studies propose to use Knowledge Graph (KG) due to its high expressivity and compatibility for representing kinds of knowledge. However, the ZSL community is still short of standard benchmarks for studying and comparing different KG-based ZSL methods. In this paper, we proposed 5 resources for KG-based research in zero-shot image classification (ZS-IMGC) and zero-shot KG completion (ZS-KGC). For each resource, we contributed a benchmark and its KG with semantics ranging from text to attributes, from relational knowledge to logical expressions. We have clearly presented how the resources are constructed, their statistics and formats, and how they can be utilized with cases in evaluating ZSL methods' performance and explanations. Our resources are available at https://github.com/China-UK-ZSL/Resources_for_KZSL.