Papers of the day   All papers

Noise or Signal: The Role of Image Backgrounds in Object Recognition

Comments

Ankur Handa: Very interesting work from @aleks_madry's lab where they find that classification networks can rely on background alone and also misclassify with adversarially chosen backgrounds. They conclude that accurate models depend less on background. https://arxiv.org/abs/2006.09994 https://t.co/CUbMSJYJoe

3 replies, 39 likes


phalanx: Noise or Signal: The Role of Image Backgrounds in Object Recognition Assessment of the tendency of state-of-the-art object recognition models to depend on signals from image backgrounds https://arxiv.org/abs/2006.09994 https://t.co/AfLkOk5FO6

0 replies, 21 likes


Aleksander Madry: Our paper is also available on arXiv: http://arxiv.org/abs/2006.09994 (3/3)

0 replies, 12 likes


arXiv in review: #NeurIPS2020 Noise or Signal: The Role of Image Backgrounds in Object Recognition. (arXiv:2006.09994v1 [cs\.CV]) http://arxiv.org/abs/2006.09994

0 replies, 1 likes


akira: https://arxiv.org/abs/2006.09994 The dependence of the classifier on the background is verified and a dataset IN-9 is provided for this purpose. They confirm the classifiers depend not only on the foreground, but also on the background and reducing the confidence by selecting background. https://t.co/NndKTaQSsU

0 replies, 1 likes


Content

Found on Jun 19 2020 at https://arxiv.org/pdf/2006.09994.pdf

PDF content of a computer science paper: Noise or Signal: The Role of Image Backgrounds in Object Recognition