Papers of the day   All papers

DIVNOISING: Diversity Denoising with Fully Convolutional Variational Autoencoders


Florian Jug: Ops, we did it again! Image denoising just leveled up!!! Our new method can do so much more than just predicting a denoised image! Let us take you on a short tour… (1/6) #CARE #denoising #uncertainty #diversity @Mangal_Prakash_ @sagzehn @jug_lab

11 replies, 377 likes

Alexander Krull: #DivNoising ( is unsupervised denoising that generates diverse solutions, accounting for the uncertainty introduced by the noise. Now, you can use it yourself! We just published our code ( Thanks to @florianjug @Mangal_Prakash_!

1 replies, 94 likes

Mangal Prakash: Ever wondered that it would be cool if image denoising methods gave you multiple plausible diverse denoised solutions for your noisy images? This would really help remove ambiguities, right? Then, check our latest preprint which uses Variational Autoencoders to do so

3 replies, 47 likes

Mangal Prakash: Our #DivNoising code is public now. Try it out, let us know how well it works for your data

1 replies, 40 likes

Alexander Krull: Why denoise your images when you can DivNoise them? Why should you settle for a single estimate when you can have the full posterior? Of course DivNoising can be trained without clean ground truth. Check it out.

0 replies, 31 likes

ʝuɢ lab: Latest lab preprint. Makes us really proud! Great team effort!!! 🎉

1 replies, 30 likes

Florian Jug: Bonus: the networks we used in the manuscript are tiny, most of them fitting in 2GB of your GPU. Hence, if you give us some time your Fiji might start talking #DivNoising to you... 😊 @FijiSc

2 replies, 19 likes

Florian Jug: A #DivNoising network is trained to know how structures in the data look like and can be used to predict diverse denoised outputs. In areas of the image where the input is ambiguous, #DivNoising will sample the space of reasonable interpretations.(3/6)

2 replies, 15 likes

Florian Jug: With #DivNoising we propose a fundamentally new way to train neural networks for image denoising, using VAEs. No additional training data needed, only noisy data is enough! The required noise model can also be bootstrapped from the noisy input data… (4/6)

1 replies, 7 likes

Simon F. Nørrelykke: Yet another must-read from the @jug_lab: Creating a distribution of denoised images, that in turn could lead to a distribution of possible segmentations. #BioImageAnalysis #DeepLearning

0 replies, 6 likes

Alister Burt: Wow - congratulations, really looking forward to reading!

0 replies, 4 likes

Ulugbek S. Kamilov: Nice work on pushing the limits of image denoising by Florian Jug. Image denoising is fun; “Regularization by Artifact Removal” is the way to push these ideas to the whole “imaging inverse problem” world

1 replies, 3 likes

Florian Jug: Check the preprint to see that this all makes sense also from a more formal point of view. We are really excited about this and are sure that #DivNoising is powerful! We decided to demo its usefulness on an instance segmentation task. (5/6)

1 replies, 2 likes

Ben Engel: #DivNoise uses fully-convolutional variational autoencoders to produce diverse denoised outputs, and hopefully an improved consensus. It requires no other training data, and can bootstrap its own noise model from the noisy data 🤩. @florianjug, does it work for #cryoET data?❄️🔬

1 replies, 2 likes

Florian Jug: What always bugged us with any CARE method is that a single restored output is created, giving the false impression that the input can unambiguously be restored. But some information gets lost in the noise, so this can obviously not strictly be true! (2/6)

1 replies, 2 likes

Erik Olsen: Something for de-noising otolith images? @nilsolavh @Tartificiel

0 replies, 1 likes

HotComputerScience: Most popular computer science paper of the day: "DIVNOISING: Diversity Denoising with Fully Convolutional Variational Autoencoders"

0 replies, 1 likes


Found on Jun 12 2020 at

PDF content of a computer science paper: DIVNOISING: Diversity Denoising with Fully Convolutional Variational Autoencoders