Arun Mallya
@arunmallya
Followers
603
Following
47
Media
154
Statuses
274
Explore trending content on Musk Viewer
باريس
• 825612 Tweets
LINGLING MAJOR FANDOM
• 649744 Tweets
Colorado
• 209290 Tweets
Brazil
• 207364 Tweets
#انقاذ_النصر_مطلب_عشاقه
• 201170 Tweets
ليفربول
• 194483 Tweets
بايرن
• 188692 Tweets
بنفيكا
• 187139 Tweets
انتر
• 186879 Tweets
دورتموند
• 186458 Tweets
ليفركوزن
• 184754 Tweets
#UCLdraw
• 109380 Tweets
Bayern
• 61135 Tweets
Your Season On Stage
• 60041 Tweets
Monaco
• 41888 Tweets
#ابعاد_لودي_مطلبنا
• 40740 Tweets
Atalanta
• 25745 Tweets
Dortmund
• 22302 Tweets
Leipzig
• 22069 Tweets
Leverkusen
• 21709 Tweets
Jabar
• 19525 Tweets
Brest
• 19117 Tweets
Anfield
• 18943 Tweets
Girona
• 17896 Tweets
#ChampionsLeague
• 14963 Tweets
ハチャウマ
• 14898 Tweets
Jurassic World
• 13393 Tweets
Mason Mount
• 10203 Tweets
Introducing GANcraft, a method to convert user-created semantic 3D block worlds, like those from Minecraft, to realistic-looking worlds, without paired training data!
arxiv:
webpage:
by
@zekunhao19951
,
@SergeBelongie
,
@liu_mingyu
11
217
794
Code for
#GANcraft
(ICCV'21) has been released at , with pretrained models & training instructions. You can even import your own worlds and make them real!
This also includes updates to the
#Imaginaire
repo to make it faster, better, and more awesome!
3
53
292
Learn how to speed up your neural network training by 2-15x with no loss in accuracy! 🤯🤩
The new and updated "Accelerating Computer Vision with Mixed Precision" tutorial is online as part of
#ECCV2020
.
Ft. talks by
@liu_mingyu
@tcwang0509
@shalinidemello
3
41
158
Thanks for everything.
“How lucky I am to have something that makes saying goodbye so hard.” - Pooh Bear to Pinglet.
5
1
68
GANcraft also allows you to condition the outputs on a style-conditioning image, and interpolate between styles. It works on large input worlds of 512x512x256 blocks, corresponding to 32 real-world soccer fields! It still produces features smaller than a block, e.g. flowers!
1
7
57
Do Deep Networks dream of (Electric) Sheep?
Yes, they do!
Introducing
#DeepInvert
(). It allows the synthesis of previously used training data, given just a trained model.
*No new model training, no training metadata required.*
2
15
41
An update to
#Imaginaire
The pretrained world-consistent vid2vid () model for the MannequinChallenge dataset has been released!
Code and a lot of other models at:
By
@liu_mingyu
@tcwang0509
@xunhuang1995
0
8
35
In the absence of paired training data mapping Minecraft scenes to real images, we use "pseudo-ground truths", generated by a GAN pretrained image-to-image translation network. These serve as surprisingly good proxies to real ground truths.
2
2
25
It can handle diverse input labels in the same world, for example, water, grass, trees, and snow covered mountains in this scene (input 3D block world shown in inset).
1
0
20
It can even handle input worlds with label distributions very different from collections of real-world imagery. For example, in the video below, the desert label accounts for > 50% of the world, while online landscape photo collections typically have a lower fraction of desert.
1
1
20
As im2im models are not view-consistent, we try to explain away inconsistencies via learning a style encoder that modulates generator activations. By using a combination of pixelwise, perceptual, and adversarial losses, we obtain realistic-looking outputs.
1
0
14
"But this leads me to a piece of advice I really hope we all consider. To all the young scholars out there in AI: You should finish your PhD with three papers that you are decidedly passionate about. Three papers that you can tell a strong story about."
2
3
15
Lastly, a plug for , which supercharged this project with its mixed-precision training and collection of useful GAN models, layers, and losses!
0
0
15
Happy to announce that all the papers I submitted to
#NeurIPS2021
have been accepted!
Sadly, the acceptance rate of my submissions is undefined 😑
1
0
14
A video with more results and method overview is available at
1
1
11
#GANcraft
is being presented as an oral paper at ICCV 2021.
Details here:
Paper website:
Work in collaboration with
@zekunhao19951
,
@SergeBelongie
, &
@liu_mingyu
at
@NVIDIAAI
.
Check out
#GANcraft
during Paper Sessions 11A and 11B: Wednesday, October 13, 6:00 PM – 7:00 PM EDT & Friday, October 15, 11:00 AM – 12:00 PM EDT
Time zone convertor:
Paper website:
1
0
6
0
5
10
We cover mixed precision basics, PyTorch code optimizations, & training of GANs, LSTMs, 3D networks, and more!
Watch it all at
0
1
7
SPACE🚀: Speech-driven Portrait Animation with Controllable Expression will be featured at
#ICCV2023
!
(02:30 pm - 04:30 pm POSTERS @ Room Nord)
Come meet us!
SPACE generates high-resolution, expressive videos with realistic head pose, using just speech and a single image!!
1
2
6
Check out
#GANcraft
during Paper Sessions 11A and 11B: Wednesday, October 13, 6:00 PM – 7:00 PM EDT & Friday, October 15, 11:00 AM – 12:00 PM EDT
Time zone convertor:
Paper website:
1
0
6
Stop by for Dreaming to Distill tomorrow!
Session: Oral 2.4C, Transfer/low-shot/unsupervised learning, 1st slot.
Window 1: Wed. Jun. 17, 16:00-18:00pm PDT
Window 2: Thu. Jun. 18, 4:00-6:00am PDT
Paper:
Code:
0
3
5
🎉 The hottest Getty + NVIDIA collab just dropped! 🎉
My colleagues and I, part of Deep Imagination Research, have been hard at work to bring you the best commercially safe image generator. Check it out!
0
1
4
0
1
4
With the new OpenAI is not Open extension, every instance of OpenAI will be replaced by "Open"AI.
What's in a name, anyway?
¯\_(ツ)_/¯
0
0
3
The reward for surviving the Half Moon Bay Road descent in traffic!
0
0
3
First in Nov 2020, now again. New company, same old trail, same old pain. Didn't need a bike light to see in the dark, did see a snek.
0
0
3
Deep Imagination Research at NVIDIA has a new website and we're hiring for roles in generative content creation -
0
0
3
Really proud of
@MridulaMallya
for founding to help you adopt stray/rescued dogs in Bangalore.
Adopt, don't shop!
1
0
3
@jbhuang0604
All we need to do is backprop through this network and find new research topics!
Might just devolve into a GAN-setting ultimately...
1
0
3
"The term was coined in Neal Stephenson's 1992 science fiction novel Snow Crash, where humans, as avatars, interact with each other and software agents, in a three-dimensional space that uses the metaphor of the real world." - Took 29 years, but the Metaverse is here!
QUIET ON SET!🎬
In just over a week, the global documentary premiere of "Connecting in the Metaverse: The Making of the GTC Keynote" goes live at
#SIGGRAPH2021
at 11am PT on August 11.
The countdown begins 👉
6
36
128
1
0
2
Partial (annular) solar eclipse captured from the Sierra Nevada!
0
0
2
Based on
@zzznah
’s
#DeepDream
, our method performs intermediate feature regularization using stored batchnorm statistics.We are able to generate all previously seen training classes with unmatched realism and fidelity.
1
1
2
Just Bay Area license plates...
I can confirm both husband and wife are PhDs 😃
0
0
2
She was brutal, cruel, and mean. Freezing winds, washed out rocky trails, harsh sun, and high altitudes. But I'll live to summit another day. All thanks to my partners in misadventure.
0
0
2
These synthesized dream images enable a new class of "data-free" tasks which do not need any prior images/labeled data/metadata.
We show applications to knowledge transfer, network pruning & incremental learning.
This work poses new questions about data & model privacy/security.
1
0
1
@phrogz42
Having paired training data would definitely improve the realism, but I imagine it would be hard to scale up skilled manual effort. The "pseudo-ground truths" were a way to transfer knowledge from powerful im2im translation networks to this task.
0
0
1
In summary:
- SPACE is a powerful tool for animating a facial image using just speech.
- Offers controllability of head pose, emotion label and intensity, blinking, and eye gaze control.
- Produces photorealistic and temporally stable outputs
Learn more at
0
0
1
Thanks to all the hard work and effort of interns - Hongxu (Danny) Yin and Zhizhong Li, as well as collaborators at
@NvidiaAI
including Pavlo Molchanov,
@ALVAREZ_JOSEM
,
@jankautz
.
0
0
1
SPACE decomposes the task of speech to face animation into 3 stages: 1) Speech2Landmarks (S2L), 2) Landmarks2Latents (L2L), and 3) Video Synthesis. This allows for any keypoint transformations, including blinking and eye gaze
1
0
1
We are never finished with grief. It is part of the fabric of living. It is always waiting to happen. Love makes memories and life precious; the grief that comes to us is proportionate to that love and is inescapable.
0
0
1
@rahuld3eora
@jankautz
@liu_mingyu
@tcwang0509
@shalinidemello
No, it in fact improves performance. Here are some slides on segmentation from the first edition of this workshop held at ICCV 2019 -
Also, code for Mask-RCNN -
and other seg. methods -
1
0
1
Yes! More memes please!
0
0
1
"But now that he is what he is, he thinks of her all the time, and the best thing about life and the worst is that she exists."
"The illicit happiness of others" has been a frustrating read at times but drops gems at unexpected times
0
0
1
@ArashVahdat
@StackOverflow
Hah, I lost track of the date and didn't realize it was a joke. Thought they made a really bad business decision 😂
1
0
1
This is great advice. But sadly, it isn't easy to follow especially when facing the publishing/arxiv/blog post firehose.
0
0
0
Hoover damn that's very little water. 150 ft drop from the 80s. Climate change is here and we're too late.
0
0
1
Get lit with Lumos!
Learning to Relight Portrait Images via a Virtual Light Stage and Synthetic-to-Real Adaptation
abs:
project page:
3
23
131
0
0
1
@ArashVahdat
@yisongyue
2020 is more of a mood, not just a calendar year 😂😭
2020 will probably continue for another half year
0
0
1
According to Inyo National Forest, over 16,000 people applied for permits to hike Mt. Whitney. Each applicant paid a $6 application fee to be considered in the lottery. That money from application fees goes to Booz Allen Hamilton (not NPS)😢!
2
0
1
