Overview of recently read CV articles

[1] Ejaz Ahmed, Michael Jones, Tim k.marks. An improved deep learning Architecture for person re-identification. In CVPR 2015.

Using DNN for pedestrian recognition, two new Layer-cross-input neighborhood Differences layer, Across Patch Features layer were introduced. The former is used to compare the input of two feature maps. and two kinds of "differences" are calculated according to the defined formula, which is used to integrate the two differences into a higher-order feature map to measure the similarity of two graphs. Anyway, see the picture of the architecture below!

Overview of recently read CV articles