Computer Vision News - June 2019

27 Focus on Computer Vision News While t-SNE has great success and highly appealing results, you should take care of three caveats of this model: first, the clusters of the model might appear in non-cluster data due to incorrect parameters settings. Interactive exploration may thus be necessary to set the correct parameters. Second, t-SNE uses Euclidean metric, which makes it suffer from the curse of dimensionality when using high dimensional data. Lastly, the optimization process is not convex, and hence, convergence is not guaranteed . We highly recommend that you try the model yourself using our code, to check whether it is useful for you. Feedback of the Month Ilya and the team at RSIP Vision provided us with a neat computer vision solution to a multi-layered problem in our project. We also consulted with the RSIP Vision team regarding the following steps in our project, and these discussions inspired us to find the ways to tackle some very challenging computer vision tasks . Ori Weitz SW Ventures t-SNE