We thank the reviewers for mentioning a few pioneering works [8, 58, D] that learn to localize

We did not reproduce results of [C] on SUN due to time limit (will include in final paper). Here we compare our work with these papers and clarify our novelty. Besides, they can only localize a small number of object parts, i.e., [58] for 2 parts, [D] for 4 parts, and are not able to "the employment of these ideas together for attribute localization and ZSL is quite interesting and seems to lead to The discussion will be added in the final paper. APN can obtain attention maps that better localize visual attributes compared to CAM (See Figure 1 in the main paper). CNN to predict binary attributes without prototypes.