innovator

Add idea


Calendar

«    March 2019    »
MonTueWedThuFriSatSun
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31

 

Advert

 

Payment

 

Advert

 

Authorization

Стартап
DateDate: 15-04-2018, 06:19

A new program, working with ultra-precise and recurrent neural networks, is able to turn blurry images into unusual short videos, each of which can include up to 10 frames.
A unique idea to restore some movement from its artifacts on a blurred image, and turn it into a full-fledged short video arose from specialists at the Indian Institute of Technology. Under the guidance of AN Rajagopalan, scientists developed a special algorithm based on convolutional neural networks (used to automatically recognize images) and recurrent neural networks.
The essence of the technique is as follows: the model is trained on many rolls, broken into frames, after which it chooses an episode, the artifact on which most corresponds to that in the training sample. The decoder then generates the fragment artifacts in the training sample into action on the video. As a result, the neural network creates a video from a blurry photo, consisting of 10 frames.
Such an algorithm in the future can be used to improve recovered blurry pictures and video.
Source: VistaNews





Add comment

Your name:*
Your E-Mail:*
Input code
from the picture*
Enable this picture to display the security code
reload, if the code cannot be seen