engleski

Natural image understanding: principles, challenges and research outlook

Development of supervised convolutional models has resulted in tremendous advances of visual recognition. Today we are able to design systems which classify images, detect faces and recognize people at a near human level. However, challenges such as cross-dataset generalization or adversarial examples indicate that further work is required to reach truly intelligent artificial vision. In this talk I will present current state of visual recognition: operational principles, current performance, main challenges, as well as research directions which promise to push the limits even further.

computer vision

nije evidentirano