"Generative model of fonts as SVG instead of pixels. Structured format enables flexible manipulation"
🔎 arxiv.org/abs/1904.02632
📉 @loss_function_porn
🔎 arxiv.org/abs/1904.02632
📉 @loss_function_porn
This media is not supported in your browser
VIEW IN TELEGRAM
"HoloGAN: Unsupervised learning of 3D representations from natural images"
🔎 arxiv.org/abs/1904.01326
📉 @loss_function_porn
🔎 arxiv.org/abs/1904.01326
📉 @loss_function_porn
This media is not supported in your browser
VIEW IN TELEGRAM
"New milestone in style transfer: direct optimization of brushstrokes via backpropagation "
👤 @ReiiYoda (twitter)
📉 @loss_function_porn
👤 @ReiiYoda (twitter)
📉 @loss_function_porn
"AI music band! Lyrics: GPT-2, composition: hierdec-trio_16bar, vocals: Synthesizer V, visuals: Sketch RNN"
🔎 youtu.be/UKF_eDyq-B8
📉 @loss_function_porn
🔎 youtu.be/UKF_eDyq-B8
📉 @loss_function_porn
YouTube
AI Sheep Song (Using GPT-2, Synthesizer V, and Google Magenta)
This song was written, composed, and sung by several narrow AI technologies in about 4 minutes.
LYRICS: Sample output from Open AI's GPT-2 when conditioned on text from the wikipedia page for sheep
https://github.com/openai/gpt-2
COMPOSITION: Six concatenated…
LYRICS: Sample output from Open AI's GPT-2 when conditioned on text from the wikipedia page for sheep
https://github.com/openai/gpt-2
COMPOSITION: Six concatenated…
"MuseNet, new GPT-2-based network by OpenAI, gives a concert in 2 minutes!"
🔎 openai.com/blog/musenet
🔎 stream: www.twitch.tv/openai
📉 @loss_function_porn
🔎 openai.com/blog/musenet
🔎 stream: www.twitch.tv/openai
📉 @loss_function_porn
"Using DeepFakes for visual translation of commercial to 9 languages. In the very near future all movies' dubbing will be fully synchronized with lips. Wow!"
🔎 youtu.be/QiiSAvKJIHo
📉 @loss_function_porn
🔎 youtu.be/QiiSAvKJIHo
📉 @loss_function_porn
YouTube
David Beckham speaks nine languages to launch Malaria Must Die Voice Petition
David Beckham launches the world's first ever voice petition to end malaria. Join him and speak up. Go to malariamustdie.com and say 'Malaria Must Die'.
This media is not supported in your browser
VIEW IN TELEGRAM
"I'm very proud to share our first paper Learnable Triangulation of Human Pose. We smashed previous state of the art (by ~2.5x) in multi-view 3D human pose estimation. Like, share, cite!"
🔎 saic-violet.github.io/learnable-triangulation
🔎 arxiv.org/abs/1905.05754
📉 @loss_function_porn
🔎 saic-violet.github.io/learnable-triangulation
🔎 arxiv.org/abs/1905.05754
📉 @loss_function_porn
This media is not supported in your browser
VIEW IN TELEGRAM
"One more paper from our lab. Authors animate talking heads with facial landmarks using only few shots of target person (or even 1 shot). Revive Dostоevsky!"
🔎 youtu.be/p1b5aiTrGzY
🔎 arxiv.org/abs/1905.08233
📉 @loss_function_porn
🔎 youtu.be/p1b5aiTrGzY
🔎 arxiv.org/abs/1905.08233
📉 @loss_function_porn
This media is not supported in your browser
VIEW IN TELEGRAM
Predicting depth of moving people captured with moving cameras. Trained on YouTube Mannequin Challenge videos.
📝 arxiv.org/abs/1904.11111
📉 @loss_function_porn
📝 arxiv.org/abs/1904.11111
📉 @loss_function_porn
The model in the center is 100% computer rendered Instagram model. #nolearning
🔎 instagram.com/imma.gram
👤 @hackermaderas (twitter)
📉 @loss_function_porn
🔎 instagram.com/imma.gram
👤 @hackermaderas (twitter)
📉 @loss_function_porn
This media is not supported in your browser
VIEW IN TELEGRAM
And again paper from our lab (Samsung AI Center, Moscow). Neural free-viewpoint rendering of human avatars. Accepted to CVPR'19.
🔎 saic-violet.github.io/texturedavatar
📝 arxiv.org/abs/1905.08776
📉 @loss_function_porn
🔎 saic-violet.github.io/texturedavatar
📝 arxiv.org/abs/1905.08776
📉 @loss_function_porn
This media is not supported in your browser
VIEW IN TELEGRAM
Novel Sparse Transformer is able to maintain a much larger context with unlearnable sparse connectivities. Looks like we are about to have a breakthrough in generation of music and text!
🔎 openai.com/blog/sparse-transformer
📉 @loss_function_porn
🔎 openai.com/blog/sparse-transformer
📉 @loss_function_porn
This media is not supported in your browser
VIEW IN TELEGRAM
CVPR is coming! Here's a hot submission on speech-to-gesture prediction. Does it have an Italian mode?
▶️ youtu.be/xzTE5sobpFY
🔎 people.eecs.berkeley.edu/~shiry/projects/speech2gesture
📉 @loss_function_porn
▶️ youtu.be/xzTE5sobpFY
🔎 people.eecs.berkeley.edu/~shiry/projects/speech2gesture
📉 @loss_function_porn
This media is not supported in your browser
VIEW IN TELEGRAM
Neural Rendering in the Wild:
1. Take ~3K Internet photos
2. Retrieve 3D point cloud from photos using any structure from motion algorithm
3. Enjoy realistic free-point-of-view video, rerendered by NN
🔎 arxiv.org/abs/1904.04290
📉 @loss_function_porn
1. Take ~3K Internet photos
2. Retrieve 3D point cloud from photos using any structure from motion algorithm
3. Enjoy realistic free-point-of-view video, rerendered by NN
🔎 arxiv.org/abs/1904.04290
📉 @loss_function_porn
This media is not supported in your browser
VIEW IN TELEGRAM
DeepView: generate new views having sparse set of input images from different viewpoints
🔎 arxiv.org/abs/1906.07316
📉 @loss_function_porn
🔎 arxiv.org/abs/1906.07316
📉 @loss_function_porn