This is interesting but suffers from not being written by native speakers. For example this section is unclear "Firstly, we obtain a visual memory by aggregati

This is interesting but suffers from not being written by native speakers. For example this section is unclear "Firstly, we obtain a visual memory by aggregating the knowledge from both head and tail classes. Then the visual concepts stored in the memory are infused back as associated memory feature to enhance the original direct feature. It can be understood as using induced knowledge (i.e. memory feature) to assist the direct observation (i.e. direct feature)." What is a "visual memory?" Are they talking about a learned feature space? An architecture isn't even mentioned. The word "infused" is used repeatedly in the post but isn't well defined. Do they mean that there is a concatenation of vectors? Or, from the diagram, is there an attentional weighting mechanism in play? I'm sure I'll understand more when I read the paper, but this blog post doesn't have the intended effect reaching a broader audience through the use of plain language.