Creating and sharing GIFs is fun. Apart from emojis, GIFs are becoming the favorite choice for the people to express their feelings while in middle of a conversation on a plethora of chatting apps like Messenger, Telegram, Whatsapp and Snapchat. Some popular apps like Whatsapp provide an easy access to thousands of GIFs owing to the GIPHY integration within the app.
Other than that, there are a number of websites including mr-gif, Gifbin and Littleplasticthings. But the thing is, you’ll find most of the GIFs in pretty bad shape. This is primarily because of the availability of enormous content over the web. Since most of the GIFs are created from the video clips, they are compressed to extremely small size which in turn affects their quality. I know what you might be thinking by the time. It’s easy to degrade the quality of an image or video but it seems like an impossible task do the opposite. Well, with Machine Learning, everything’s possible.
How ML Will Fix This?
There’s a similar website called Gfycat that’s working on solving this issue by leveraging the innovative aspects of Machine Learning. The company is working on the technical front to make the low-fidelity GIFs look good by improving their quality. The website features a wide variety of GIFs in every category including sports, entertainment, fun, dance, celebrities, mood, expressions and emotions.
As per the company CEO, Richard Rabbat, the problem with these GIFs is mainly related to the video and is less to do with the image recognition. For this reason, Gfycat will look for the original video source across the web from which the GIF has been created. Most of these videos can be found on YouTube.
The app will analyse the entire video to find out what part of the video has been turned into that particular GIF. Gfycat will then produce a high-quality GIF from that video and will swap it with the low-quality GIF to make it look better.
The Challenges
The biggest challenge is perhaps creating a high-quality GIF out of a shoddy GIF featuring a celebrity or a public figure. As a GIF is uploaded by a person, he seldom tags that celebrity while making the upload. So it gets difficult to identify which celebrity is featured in the GIF. For this, the company has build up a vast library of popular celebrities. With this, the company hopes to identify the celebrity GIFs coming in and tag them automatically.
Another big challenge would be to identify and replace the text shown in the GIFs as caption. In most of the GIFs that we see, the text is grainy and not easy to read. For this, Gfycat aims on building some internal tools that would help identify what the captions say and then machine learning will rewrite the caption with clear text.
Clearly, Gfycat is not the only one with such an endeavor. Several other similar companies like Tenor and Giphy are also working in this direction. Let’s hope things work out well for these companies and we get a robust platform for refurbishing the low-fidelity GIF images.
Read more info at — https://www.oodlestechnologies.com
No comments:
Post a Comment