Upgrade to Pro — share decks privately, control downloads, hide ads and more …

機械学習の現在とPython

masa-ita
January 19, 2019

 機械学習の現在とPython

masa-ita

January 19, 2019
Tweet

More Decks by masa-ita

Other Decks in Technology

Transcript

  1. • • Video-to-Video Synthesis Figure 1: Generating a photorealistic video

    from an input segmentation map video on Cityscapes. Top left: input. Top right: pix2pixHD. Bottom left: COVST. Bottom right: vid2vid (ours). Click the image to play the video clip in a browser. given paired input and output videos, With carefully-designed generators and discriminators, and a new spatio-temporal learning objective, our method can learn to synthesize high-resolution, photore- alistic, temporally coherent videos. Moreover, we extend our method to multimodal video synthesis. Conditioning on the same input, our model can produce videos with diverse appearances. We conduct extensive experiments on several datasets on the task of converting a sequence of segmentation masks to photorealistic videos. Both quantitative and qualitative results indicate that our synthesized footage looks more photorealistic than those from strong baselines. See Figure 1 for example. We further demonstrate that the proposed approach can generate photorealistic 2K resolution videos, up to 30 seconds long. Our method also grants users flexible high-level control
  2. • • • 128x128 GAWWN 256x256 StackGAN Text description 64x64

    GAN-INT-CLS This small bird has a white breast, light grey head, and black wings and tail A bird with a medium orange bill white body gray wings and webbed feet A small yellow bird with a black crown and a short black pointed beak A small bird with varying shades of brown with white under the eyes The bird is short and stubby with yellow on its body This bird is red and brown in color, with a stubby beak This small black bird has a short, slightly curved bill and long legs Figure 3. Example results by our StackGAN, GAWWN [24], and GAN-INT-CLS [26] conditioned on text descriptions from CUB test set. Han, Z. et. el, StackGAN: Text to Photo-realistic Image Synthesis with Stacked Generative Adversarial Networks
  3. • ' -4 :<7? 56 I- • &.4 &-4 •

    ,4 &0A @ • KE =C4-) • RG(L-)6I8$ *<2>%RG(L :< 4;-) • H/4 N#MO -QPJF-) &6I-)" • (B-) &" • 19!D3 4;-)+ " https://leapmind.io/blog/2018/12/26/2018-ai-news/
  4. • >? BC+-*F • ./&M N, 0KJ2+- • (.@ %9184

    • ). 7: GH6). E;+-.DLO 5<! • / '#="/ $ 3I A '#=" https://leapmind.io/blog/2018/12/26/2018-ai-news/
  5. mcgill tum.de inria.fr in.tum.de ualberta.ca uic nvidia.com cse.iitm.ac.in nus rice

    kaist.ac.kr caltech de.bosch.com ttic ntu hotmail.com amazon.com umass uci tuebingen.mpg.de umd imperial.ac.uk salesforce ucla duke socher.org uw ucsd uva.nl upenn princeton ed.ac.uk baidu.com openai.com cam.ac.uk umich utexas ethz.ch usc columbia illinois washington pku cornell toronto gatech oxford nyu tsinghua montreal ibm facebook mit stanford microsoft.com berkeley cmu google.com Top 100 prolific organizations Oral Poster Workshop Reject https://prlz77.github.io/iclr2018-stats-3/    Toyota Technological Institute at Chicago