OPPO FDF full-dimensional portrait video technology system debuts at INNO DAY

November 18, 2020, Shenzhen, China—Today, OPPO launched the OPPO FDF full-dimensional portrait video technology system for the first time at the Future Technology Conference 2020 (OPPO INNO DAY 2020) with the theme of "Leap Transition and Goodness". The release of OPPO's accumulation of beauty technology over the years is an evolution in the field of portrait aesthetics and an innovation in the 5G video era.

Set a new benchmark for all-scene video beauty in the 5G era

Image Product Manager Tian Zhijian said, “Image has always been the track that OPPO insists on and will stick to for a long time. We always insist on starting from the user in the definition of image products. In order to allow users to shoot natural, confident, and moving portraits, we have launched magic internally. Mirror project, in different cities and Reno series old users, terminal shopping guides, portrait finishers and makeup experts in-depth discussion of the portrait videos they think are good. Unexpectedly, almost everyone thinks that good portrait videos Not only to achieve beauty, but also beauty. Based on this insight, we have created the OPPO FDF full-dimensional portrait video technology system. Of course, this is also the concept we have been adhering to in portrait shooting-let beauty grow naturally."

The OPPO FDF full-dimensional portrait video technology system is composed of two engines: portrait perception and image quality enhancement, which are respectively optimized for the subject of the character and the overall picture during portrait shooting, and jointly achieve the beauty of people and scenery, whether it is a single person It is still multi-person, day, night or backlight, you can achieve full-time video creation freedom.

Reject the sameness and make beauty more natural

The perceptual portrait engine has AI face detection, AI super clear portrait, OFL skin quality optimization algorithm, facial distortion correction and other functions, rejecting the same thing, making the shots more natural, real and moving.

AI face detection achieves 99.78% detection accuracy and 25ms single frame detection time by optimizing the network model. Whether it is shooting video or taking pictures, it can quickly detect the face area and optimize the delineation for subsequent faces The precise range. At the same time, the AI algorithm marked 296 key points in the face area to distinguish the facial features and contour details, and added auxiliary information such as whether the key points are visible or not, the face angle judgment, etc., and finally achieved a low level on the industry's authoritative WFLW data set. To 4.71% of NME, the frame rate of this algorithm on mobile platforms has reached more than 200fps.

The AI super-clear portrait has designed 3 sub-modules in the deep neural network model to perform super-resolution calculation, noise reduction optimization and sharpness optimization on the face. Even in a low light environment, the algorithm can accurately detect the face and neighboring areas and perform enhancement processing to enhance the clarity of the face while making the details more refined.

The OFL skin quality optimization algorithm can perform AI processing on skin color and skin quality separately, helping users remove facial blemishes, optimize skin performance, and make the overall complexion more outstanding during shooting. This algorithm can segment 96.5% of the skin area, and the accuracy of facial skin reaches 97.4%. "Skin color detection" includes not only facial skin, but also neck and hand skin, covering almost all skin areas that will be exposed when shooting portrait videos.

The OFL skin quality optimization algorithm can also achieve the AI blemish removal effect, intelligently distinguish the characteristics of natural and acquired blemishes, and achieve a more natural blemish removal effect. Acne spots, acne, bags under the eyes and other acquired blemishes will be removed, and natural facial features such as moles will be preserved.

When dealing with face distortion, it takes the background part in the perspective projection, cuts and separates the face area, and then performs a spherical projection to keep the face in a normal shape that conforms to the viewing habits of the human eye. Superimpose the two images to get a normal face and background image, making the overall effect more natural and real.

More freedom of creation in all-scene and all-weather shooting

The image quality enhancement engine includes a new generation of super anti-shake algorithm, mobile time-lapse photography, super night scene video, DOL-HDR and other functions, which can guarantee a satisfactory picture in scenes such as large picture shake, insufficient light/over bright and other scenes.

The super anti-shake algorithm is added to the scene detection process, by capturing the data of the gyroscope, accelerometer and sensor to judge the motion trajectory, posture and exposure respectively, so as to distinguish different scenes and match different anti-shake strategies.

For handheld mobile time-lapse photography, OPPO adds motion trend judgment to the super anti-shake algorithm, which can not only make the transition between the extracted multi-frame pictures smoother, but also achieve a longer span of anti-shake, so that the movement is delayed The real shooting performance of photography almost reaches the effect of PTZ.

Super night scene video is integrated with the HDR algorithm based on the previous generation algorithm, and the brightness adjustment curve is distinguished and debugged. The algorithm first intelligently judges whether to enable the night scene or the HDR brightness adjustment curve according to the current brightness environment to achieve a more transparent and better look Video effects.

The DOL-HDR solution solves the user's pain points when shooting against the light. The long and short exposure frames are carried out almost at the same time, which greatly reduces the interval between the front and back frames, greatly improves the ghosting problem, and achieves better HDR video shooting effects.

Open sharing, the first product came out in December

Since the photo era, OPPO has invested a lot of research in the field of beauty. After several years of iterative upgrades, OPPO has applied the accumulated technology to the video field, only to allow users to shoot natural, confident, and moving portraits, so that beauty can grow naturally. . It can be said that the OPPO FDF full-dimensional portrait video technology system is the key to opening the 5G portrait video era, and will set a new benchmark for full-scene video beauty in the industry.

The first new product equipped with OPPO portrait video technology system will come out in December. OPPO will also adhere to the spirit of technology sharing and technology inclusiveness. At present, OPPO's technologies in HDR, night scene, super anti-shake and other aspects have been gradually opened to third-party platforms such as Douyin and Kuaishou, empowering ecological partners and allowing more people to enjoy To the charm of technology.