05/11/2022 huaweiphone 478 Views

P20 Pro's AI feels a little overdone

It may not be necessary to reiterate here, but the camera of the HUAWEI P20 Pro HW-01K has less noise even in dark places, and just by pressing the shutter, you can get a beautiful picture. Some people have different tastes for the flashy colors, but hearing the voices of users other than myself, the evaluation is generally high. Some people say that compact digital cameras are no longer needed.

I like the finish of the night view, and I feel that the frequency of releasing the shutter has increased more than before. In the sense that it makes me want to shoot, it can be said that latent needs have been drawn out. There weren't many smartphones like that, so I appreciate that point. As a result, the number of photographs is increasing.

On the other hand, the honest evaluation is that the P20 Pro's selling point, "AI," should be handled a little more. AI is already consumed like a buzzword, and these days every manufacturer often pushes AI. I've heard many presentations about the camera being equipped with a scene discrimination function, probably because it's an easy-to-understand and easy-to-implement application example. The "Kirin 970" built into the P20 Pro is equipped with an "NPU" that performs machine learning processing, and is said to be able to quickly recognize scenes in photos.

AI determines the scene. Automatically switches to the mode that matches the detected scene

P20 Pro with overkill AI

Certainly, the scene discrimination speed is fast, and the accuracy is relatively high. This point is convenient. However, there is a tendency that the processing after discrimination is a little too bold. On a normal terminal, scene discrimination is used for color adjustment and the like. It's to the extent of increasing the warmth when you know it's cooking, or reducing camera shake at night. However, the P20 Pro changes the mode itself with AI.

I thought this was a little... when I was shooting a slide at a presentation. At a presentation where the speaker was speaking with a slide in the background, I was trying to take a picture that included both the speaker's whole body and the slide. AI recognized this scene as "text". The problem is after that. It's true that the slide is the main part of the photo, and the recognition itself is certainly correct, but I don't like the fact that the mode has changed to "document scan". As a result, the speaker's lower body was cut off to fit the slide.

Portrait mode also activates when a person is detected. Portrait mode blurs the background and makes the person stand out. The result looks beautiful as a photograph, but sometimes it's important to know where the photo was taken, such as in a tourist spot. In this case, it is better to leave the background clear to some extent, but if the composition is centered on the person, it will be in portrait mode, so it will take some time to cancel.

Photo taken at Softbank's financial results conference. When President Son cut off suddenly, he automatically switched to portrait mode. It's nice to have the person stand out, but I wanted to keep the background as well since it's a travel destination. After all it is inconvenient when we want to take photograph to. Compared to just pressing the shutter button, the number of steps increases by one tap, which makes it easier to miss the photo opportunity. This is the reason why I wrote that AI is a bit annoying in this corner before.

When the master AI is turned off, the AI stops discriminating the scene

Therefore, I decided to turn off the "master AI" in the camera settings. Setting is easy, just press the gear icon while the camera is running and turn off "Master AI". Now the automatic mode change will not work.

On the other hand, there are also situations where it would be nice to have scene recognition by AI, such as when you want to change the color slightly when shooting food, or when you want people to be slightly more natural. I understand that you want to appeal the power of AI, but I thought it would be nice to have a mode that works a little more modestly. I felt that it would be convenient if the user could select to some extent by setting it to about three levels, strong, medium, and weak, and setting it to strong when they wanted to change the mode.