This document summarizes a research paper on multi-scale fusion subspace clustering using similarity constraints. The paper proposes a new network that uses multiple self-expressive layers at different scales. It introduces a multi-scale fusion module to combine the self-expression coefficient matrices from different layers. Additionally, it includes a similarity constraint module to guide the training of the fused coefficient matrix. The network is tested on object and face clustering datasets and compared to other subspace clustering methods. Experimental results analyze the impact of different loss functions, fusion methods, and kernel initialization approaches.