Mutiny

10 years from now there will only be GANs

Working on the #CreativitySingularity & consultant

[Mutiny @MutinyFo]

*NEW* Faceprocessor (V2) https://t.co/PavbCHCW0x improved better #facial #segmentation and augmentation new #hair segmentation module ready deployable #deepneuralnetworks #onnx #pytorch #mxnet for applications in #imageediting #imageprocessing #videoediting and similar.

[Mutiny @MutinyFo]

Released free #OpenFX plug-in for automatically detecting and tracking your onscreen talent (https://t.co/m7zmR4dOjC) works in #Nuke #AssimilateScratch BMD Fusion and #Natron #VFX #colorgrading #videoeditor
Video https://t.co/naRrl59n1W

[Mutiny @MutinyFo]

@poolio @catamorphist @avdnoord @OriolVinyalsML Had not much love with VQ-VAE-2 results on my own dataset where poor maybe because it hadn't enough data in it.

[Mutiny @MutinyFo]

Hurray we've reached another major milestone with our #imagesegmentation #dataset https://t.co/xYMKVjZSe9 #50k #DeepLearning #vision

[Mutiny @MutinyFo]

@YSongStanford FID score look good even though CIFAR-10 maybe not the most exciting one for FID scores. What bothers me in the whole paper not a single word about the hardware setup and training time. FID is all good and well but if it needs 10 times longer than StyleGAN this is not a real win.

[Mutiny @MutinyFo]

Another major update (the second one this year) on our #facesegmentation #dataset for #DeepLearning and #deepneuralnetwork this year! https://t.co/VmWxQSBNUZ

Also 41637 annotated image pairs now! Have a look and why not become a Patreon!

[Mutiny @MutinyFo]

@patrick_oshag Kerbal Space Program it is more than just a video game

[Mutiny @MutinyFo]

@mark_riedl @gradientpub I find it more astonishing that people are puzzled by this? Most of these labs are cooperate labs not public or charities. Of course they have a commercial interest to build an edge above anyone else. Don't believe those mission statements they are windowdressing most of the time

[Mutiny @MutinyFo]

@CVCND @kdnuggets Is this some kind of historical look back? This is like two years ago a lot of those top 13 are no longer continued or pretty abandoned

[Mutiny @MutinyFo]

Interesting read for sure many good arguments but would be bad for the whole field and research if it really comes down to just scale

[hardmaru @hardmaru]

My AI Timelines Have Sped Up

Interesting blog post by Alex Irpan where he describes his forecasts about when AGI could happen. A fun read, even if you may not agree with every point.

https://t.co/qUrMmaaJ1E

[Mutiny @MutinyFo]

@jm_alexia @YSongStanford If it is anywhere faster to train all for it, biggest problem with GAN is training time not FID score (IMHO). FID score you can always improve by being more selective on data, add more data or augmentation tricks

[Mutiny @MutinyFo]

@justice1996V @hardmaru Totally true but it makes even PixelCNN++ look efficient in comparison.

[PyTorch @PyTorch]

Stochastic Weight Averaging (SWA) is a simple procedure that improves generalization in deep learning over Stochastic Gradient Descent (SGD). PyTorch 1.6 now includes SWA natively. Learn more from @Pavel_Izmailov, @andrewgwils and Vincent:
https://t.co/1u5yRwfWXT

[Mutiny @MutinyFo]

Can your #GAN do #4k ? Our Cloud/Sky #GAN can! #generativeart #DeepLearning #CreativitySingularity

[Mutiny @MutinyFo]

And that's just one Raptor engine

[SpaceX @SpaceX]

Starship takes flight

[Mutiny @MutinyFo]

Excellent and highly competitive prices for what you'll get

[AIME @aime_bird]

AIME GPU Cloud now open for public. Rent affordable multi gpu high end servers for your deep learning training projects!

#artificialintelligence #deeplearning #gpucloud #cloudplatform #cloudinfrastructure

https://t.co/szIZEz9FsS

[Mutiny @MutinyFo]

I just published Mut1nyFace — A fast and accurate face detector https://t.co/Bv24yjn6LG #facedetection #YOLO #ONNX #facialediting

[Mutiny @MutinyFo]

Something I've been feeling since a while but great to see somebody did a proper study

[Denny Britz @dennybritz]

The authors of this paper analyzed 1,058 arXiv papers and plotted various benchmarks against the increase in compute requirements, arguing that the current progress is largely driven by more compute and may become unsustainable soon: https://t.co/joWHnH3QXx

[Mutiny @MutinyFo]

We released Part 2 https://t.co/9Thw0xlKtd in the series of use cases for FaceProcessor V2 #facial #segmentation. This time having fun using our #hairmatting deep #NeuralNetworks to change #hairstyles on photos thanks to the power of #GAN #photoediting #imagediting

[Mutiny @MutinyFo]

@YassineAlouini @paperswithcode All EfficientNet-based Object detector have been beaten by YoloV5 in terms of FLOPs and accuracy so maybe it is time to check again?