This document summarizes a research paper on visual relation detection using a Visual Translation Embedding Network (VTransE). It introduces the tasks of visual relation detection and the challenges of existing joint and separate models. VTransE is described as mapping object and predicate features into a low-dimensional space, where relations can be modeled as vector translations. The document outlines VTransE's feature extraction method using classemes, locations, and bilinear interpolation of visual features. It evaluates VTransE on two datasets, finding that the embedding idea is effective and certain features improve relation detection and knowledge transfer between objects and predicates. Overall, VTransE performs comparably to state-of-the-art visual relation models.