![]() Research on how strongly the background pixels affect the prediction of neural networks and how this problem can be remedied is very important in the context of the contemporary development of neural networks, the widespread acceptance of such solutions in the field of the production of everyday objects as well as autonomous cars in the future. The snow plow just because it is on the snowy background. This behavior of the network is not supported by humanĮxperience, because, for example, we do not confuse the yellow school bus with There is a visible correlation between the background and the wrong-classifiedįoreground object. MyĮxperiments on the natural and adversarial images datasets show that, indeed, The contribution of background pixels to the classifier predictions. Suppress the neurons surrounding the examined object and, consequently, reduce This purpose saliency maps created by the LICNN network. In my work, I analyze the above problem using for I suspect that theĬlassification of an object is strongly influenced by the background pixels on However, are far from ideal, because it often turns out that network can beĮffectively confused with even natural images examples. ( classifying) these objects - mainly using CNN networks. In recent years, neural networks have continued to flourish, achieving highĮfficiency in detecting relevant objects in photos or simply recognizing
0 Comments
Leave a Reply. |