Computer Vision News - July 2020

2 Summary Research 8 Figure 3. The primate brain represents visual objects through four networks tuned for four quadrants of object space: faces, bodies, spiky objects, and stubby objects. This is similar to the representation from the deep learning model proposed in the paper. It is worthwhile to explore the methods and results of this paper in more detail. A lot of comparisons are shown, such as between different networks (Inception, ResNet, DenseNet, GoogleNet) displaying the cumulative variance of the unit responses. The discovery of the new networks describing the visual cortex is an extremely important case for the features which share important for object recognition. There is a lot to be explored, as there are many unknown factors. For example, it is unclear whether the border between the patches are discreet or continuous, which could further prove if there are specialized cortex formations. Further research is also needed to explain if other unknown regions share a similar object space mapping. The cluster of neuron organization across the IT according to the similarity in their axis in the current unsupervised learning methods and the biological system represented in the Fig. 3 suggests that the two systems might share a similar mechanism for object recognition.