Papers of the day   All papers

Are we done with ImageNet?

Comments

roadrunner01: Are we done with ImageNet? pdf: https://arxiv.org/pdf/2006.07159.pdf abs: https://arxiv.org/abs/2006.07159 github: https://github.com/google-research/reassessed-imagenet https://t.co/Aahfuqy0P8

1 replies, 175 likes


Lucas Beyer: Are we done with ImageNet? That's what we set out to answer in https://arxiv.org/abs/2006.07159 with @olivierhenaff @__kolesnikov__ @XiaohuaZhai @avdnoord. Answer: it's complicated. On the way, we find a simple technique for +2.5% on ImageNet. https://t.co/a5FaxgmzmY

2 replies, 99 likes


Alexander Kolesnikov: Are we still making meaningful progress on ImageNet? What happens if we carefully re-annotate ImageNet val set? How to improve ResNet-50 top-1 accuracy by 2.5% by cleaning training data and using different loss function? See our new paper for the answers: https://arxiv.org/abs/2006.07159.

2 replies, 40 likes


olivierhenaff: Are we done with ImageNet? Yes: we found the original labels to no longer be the best predictors of human preferences. But no: with our new label-set, we have removed much of the biases of the original, creating a better benchmark for future research. http://arxiv.org/abs/2006.07159 https://t.co/VzFNdOzkaC

1 replies, 14 likes


Sebastian Raschka: Recent efforts to improve the ImageNet benchmark dataset: "Are we done with ImageNet?" -- https://arxiv.org/abs/2006.07159 PS: Isn't Purse vs Wallet a BE vs AE thing? So, AE's the lingua franca for ML, I assume? :P https://t.co/woFNtHN3Iy

1 replies, 13 likes


Bojan Tunguz: Very interesting paper. Yet another critical look at the nature of ImageNet labels, and how much its labeling idiosyncrasies mask or reinforce the true extent of the progress in Computer Vision in recent years. https://arxiv.org/abs/2006.07159 #datascience #ai #deeplearning #ml https://t.co/Xu00QwGJxE

0 replies, 11 likes


arXiv CS-CV: Are we done with ImageNet? http://arxiv.org/abs/2006.07159

0 replies, 7 likes


Florence Poirel: Congratulations to our colleagues @giffmana @__kolesnikov__ and @XiaohuaZhai from #GoogleResearch and @avdnoord and @olivierhenaff from @Deepmind for this work! https://www.twitter.com/giffmana/status/1272552809913823236

0 replies, 4 likes


stormtroper1721: Are we done with imagenet? Yes and no. https://arxiv.org/pdf/2006.07159.pdf

0 replies, 2 likes


Lior Sinclair: Are we done with ImageNet? Very interesting paper. Yet another critical look at the nature of ImageNet labels, and how much its labeling idiosyncrasies mask or reinforce the true extent of the progress in Computer Vision in recent years. https://arxiv.org/abs/2006.07159 https://t.co/1ooyZZfa1E

0 replies, 1 likes


Scott H. Hawley: Two new papers on rethinking ImageNet labels & processes used to create them: https://arxiv.org/abs/2006.07159 & https://arxiv.org/abs/2005.11295 ...from a purely technical standpoint. Neither mention the social & political issues involved, i.e. in the Person category. Timely! ;-) #classification https://t.co/H3rc6kflqP

1 replies, 0 likes


Content

Found on Jun 15 2020 at https://arxiv.org/pdf/2006.07159.pdf

PDF content of a computer science paper: Are we done with ImageNet?