As shown above, from 100x133PIx to 400x532PIx, what does it mean to have technology that can increase resolution by 400%, besides being visually clear?

It means:

  • Old photos, old movies, animation and other picture quality can be quickly improved, the resolution of free adjustment
  • The effect of industrial algorithms can be greatly improved with the improvement of data quality
  • You can also quickly view hd videos or pictures when the network signal is poor
  • Image transmission bandwidth and storage space will be greatly compressed, transmission costs will be greatly reduced

▲ Images from public data set [1][2][3]

Make the world more clearly is not magic, but a technique known as the super-resolution ascension (hereinafter referred to as “super”), to the low-resolution image, video, through feature extraction, feature fusion, image reconstruction and the sampling (Pix – Shuffle), restore and make up the information lost and eventually into high resolution image and video.

There are many ways to realize the super segmentation, and the peak signal to noise ratio PSNR32.53, structure similarity SSIM0.9083 video super segmentation algorithm pp-msvsr, with its excellent performance SOTA industry!

Without further ado, let’s directly look at the comparison chart of PP-MSVSR’s performance (peak-to-noise ratio, structural similarity) with other algorithms on the authoritative data set of video hyperpartition:

As shown in the figure, PP-MSVSR provides models with 1.45m and 7.4m parameters. The peak signal-to-noise ratio and structural similarity are both higher than those of other open source algorithms, while maintaining a small number of parameters. That is to say, PP-MSVSR achieves the most exquisite resolution improvement effect, but also maintains the advantage of small size, which is high performance and low cost in the real sense. Moreover, PP-MSVSR does not limit the resolution of the input video, supporting the resolution increase by 400% at a time, which makes the super customization to the extreme and gives everyone infinite imagination space.

The source code and tutorials for PP-MSVSR have also been open sourced to the PaddleGAN project. In addition to video hyperscoring, PaddleGAN also provides industry-leading single-image hyperscoring algorithms.

Better than action, hurry up Star⭐ collection start to try ~

Github.com/PaddlePaddl…

Now, let’s have a detailed look at how the PP-MSVSR superpartition algorithm solves the actual problem!

1. Change old movies from fuzzy to clear, and realize image restoration

In the field of entertainment, such as social networking, news media, e-commerce and other industries, it is often involved in video enhancement and editing, and THE performance of PP-MSVSR is really commendable.

2. Improve the accuracy of the algorithm by improving the clarity of training data

Whether it is target detection, image recognition, image classification, semantic segmentation and other CV related tasks, the quality of training data has a great impact on the final task effect. Therefore, developers can use PP-MSVSR to improve the video data at any resolution to the ideal resolution to achieve the accuracy of downstream detection, recognition, segmentation and other tasks.



▲ Images from public data set [4][5]

3. Reduce the cost of video network transmission and improve the resolution by using algorithms at the terminal

In this era of extreme pursuit of clarity, the cost of video storage and transmission is getting higher and higher. For example, the transmission cost of 4K signal is at least several thousand yuan a day. In order to reduce the cost, we must start from storage and transmission.

With the help of video superpartition, video cloud storage space can be greatly reduced, video CDN transmission bandwidth can be further reduced, and video transmission speed can be improved at the same time, which means that while storage space and transmission cost are greatly reduced, video transmission speed can also be significantly improved!

Pp-msvsr is so useful, is it complicated to use? The answer is: not at all! The resolution can be improved with just one command!

▲ Images from public data set [1]

PaddleGAN also provides a variety of cutting-edge GAN-related capabilities and applications, including image generation, lip synthesis, face fusion, etc. What are you waiting for? Get started at Star⭐!

Github.com/PaddlePaddl…

​

Baidu AI Developer Community baidu AI developer Community provides a platform for developers from all over the country to communicate, share and answer questions, so that developers can no longer “fight alone” on the way to research and development, and find better technical solutions through continuous communication and discussion. If you want to try all kinds of AI technologies and explore application scenarios, join baidu AI community immediately. All your ideas about AI can be realized here!

【 Contribute to win the grand Prize 】 — Fantastic ideas +AI technology =? 🌟 What sparks will be generated when ideas collide with AI technology? 🌟 what is the chemical reaction between unique and novel ideas and AI technology? 🌟 Participate in the event for the chance to win the grand prize, but also the chance to become a member of the AI Developer Alliance! Join us to unlock the developer Alliance more welfare bar awards here 👇👇👇👇 alba to participate in activities e-sports earphones, JINGdong card, iQiyi annual card, Baidu custom gift waiting for you to take! Participate now: [contribute to win the grand Prize] — Fantastic ideas +AI technology =?

​