May Day holiday, there are less than 20 hours! With us to welcome the May Day, there are 14 films, known as the history of the most crowded five! Xiaobian is right, “the most crowded” instead of “the strongest”.

“Chasing the Tiger and Catching the Dragon,” directed by Wang Jing and starring Louis Koo, Tony Leung ka-fai and Wu Chin-yu, ranked fourth at the pre-sale box office. Some disappointment, and seems to be expected.

Why do we always talk about Hong Kong films with thick filters, but at the same time we feel so mixed?

If you’ve ever seen a classic Hong Kong movie, you can understand. Chow Yun-fat, Jackie Chan, Jet Li’s emotional action, Stephen Chow’s funny comedy that makes people laugh until they cry, Lin Zheng-ying’s creepy and eye-opening, Lin Qingxia and Joey Wong’s amazing time…

May Day, then, make the new film at the box office to fly for a while, before that, let’s change the position, to relive those classic lens, just open huawei video APP, find “hd repair” in the film classification, whether Andy lau and brigitte Lin peak, or hair elder brother became a prisoner hard prison eldest brother, you will find the same plot, characters, has a different look and feel: The picture is clearer and less noisy than before, and it seems to flow more smoothly than before, without the “lag” characteristic of old movies.

How is this done? Today we’re going to find out how these hd movies are restored! Raise your posture while watching the movie.

What does film restoration do

Cinematic AI fixes (not the ones that cost a lot of people to fix like Titanic) include video fixes and video enhancements. Video repair is to solve the video quality problems caused by shooting, compression, transmission and storage, such as scratches, compression distortion, noise, etc.; Video enhancement is to further improve video quality by improving resolution, frame rate, dynamic range and color gamut.

Three major challenges in film restoration

At present, there are three major challenges in film AI restoration: data, algorithm and evaluation.

(1) Data challenge: generalization ability of blind repair task is improved

Film distortion often contains the above problems, that is, mixed distortion, blind repair, that is, before the repair, it is not clear what problems there are in the video, using the generalization algorithm, the problem is: how to improve the adaptability of the generalization algorithm to deal with a variety of repaired scenes.

Hybrid distortion

Intensity distortion

(2) Algorithm challenge: how to improve image quality in all aspects at the same time

Video enhancement is to make video images with higher pixels, clearer light, fuller colors, smoother color transitions and smoother moving images. This requires joint enhancement for optimization between multiple tasks.

(3) Evaluation challenge: Quality evaluation of generated images

After certain processing of Figure 1, traditional evaluation algorithms (such as PSNR, SSIM, VMAF, etc.) draw a conclusion that the original picture is Figure 2, but the subjective view is Figure 3. It shows that the traditional evaluation algorithm is inconsistent with the subjective evaluation algorithm.

Hard wide come: film repair which strong, and see Huawei cloud video cloud media AI services.

Huawei Cloud Video How to repair movies

Huawei Cloud video before and after the restoration of Huangfeihong

  • Mixed distortion repair

Huawei Cloud Video cloud pair may exist distortion problems, first predict, and then input into the AI algorithm, and then build a multi-task mixed distortion repair model, and finally repair, in detail, there are three aspects:

  • Airspace enhancement: video hypersegmentation

Hypersegmentation refers to the use of media processing technology to transform a low-resolution image into a higher-resolution image.

For most movies and TV dramas, audiences tend to pay more attention to faces and subtitles, which are important enhancement areas. Huawei Cloud video cloud based on semantic adaptive super-segmentation algorithm, taking into account the network lightweight, stability, comprehensive effect.

  • Time-domain enhancement: Interframe (FRC)

Time domain enhancement makes the connection between pictures smoother and smoother. Huawei Cloud Video cloud modifies optical flow by improving network and combining traditional methods, such as adjacent motion vectors in the reference time airspace.

  • Contrast and saturation enhancement

Existing contrast and saturation enhancement algorithms based on deep learning are difficult to achieve good results due to the lack of data. Huawei Cloud Video cloud can significantly improve the contrast and saturation of the picture by searching for the statistical rules of the image in the brightness and color space and serving as the prior input of the network.

Media AI service is a “pawn” in huawei cloud native video services. Based on the concept and technology of cloud native, Huawei Cloud has reconstructed audio and video media network and provided cloud native video service. First, it is fully cloud, and all video capabilities, from production to distribution, are run on the cloud. Support the whole industry, such as radio and television media, Internet video, communication collaboration, industry video applications, etc. Provide full process services from video production, processing, distribution and consumption. The result is improved production efficiency, distribution model change, continuous experience innovation, and safe, stable and reliable service.

Click to follow, the first time to learn about Huawei cloud fresh technology ~