Street View Video Mode AI may Rotates Photos 90 Degrees Wrongly

This is the Blue Line Street View of “Stairway of Success (出世の石段 - 愛宕神社)”, which is a famous place in Japan.
Let’s climb now. https://goo.gl/maps/awWkTTjHDSYGFL2XA

Of course, I didn’t mean to create such a ugly Street View.
I should have created the following Street View.
https://goo.gl/maps/gUh9ipSbuhJkGNdR8

Both of these two types of Street View were created by using GoThru.
And I’m using the exact same photo with the exact same coordinates assigned to both of them.
The only difference between these two is the processing within the Street View server.

Street View with the correct zenith was uploaded using GoThru’s Google Business mode.
This mode is the same function as the process of uploading a still image and performing constellation work in the old Street View app until June last year. (Of course, it’s much more sophisticated and works better than the old Street View app)

And the street view that has been tilted about 90 degrees was uploaded using GoThru’s Blue Line mode.
This mode lets the Street View server do the same thing as the Video mode of the old Street View app.
In Video mode (Blue Line mode), the user cannot determine the direction of the zenith. The AI ​​of the Street View server decides the zenith at their discretion.

――――――

And I provided these two modes with exactly the same pictures, with exactly the same coordinates.

Video mode (Blue Line mode) determines the direction of the zenith without the user’s intention. And that judgment is often wrong except in places that are horizontal and have a good view.
The article that summarizes the contents is as follows.
https://www.facebook.com/groups/GoogleStreetViewTrustedPhotographers/posts/2051894511652918/

And, in the image rotated 90 degrees due to the wrong judgment of the zenith like this time, the height difference of the stone steps is replaced with the left and right curves.
Their AI bent the Blue Line from side to side as if they were looking at the stone steps from the side, in order to match the connectivity of the freely rotated images. They are like boys who lie to lies and make excuses.
I wrote in the following article that the Street View server does such a thing for the blue line.
https://www.facebook.com/groups/GoogleStreetViewTrustedPhotographers/posts/2035862479922788/

The zenith direction determined by the AI ​​of the Street View server is determined by the content of the image.
This time, I imagine AI has determined that the brown building on the right side of the stone steps is the ground.

――――――

All I can do with this 90 degree tilted Street View is to delete the entire tour.
I also know that Street View servers continue to make various changes to existing images over the years.
Hoping it would be processed correctly someday, I decided to leave this Street View as it is.

6 Likes

Did you turn off the gyro?

1 Like

No.
The image I sent to the Street View server is an equirectangular with the right top and bottom.
And in video mode, the top and bottom of the photo sent by the user is ignored.
I uploaded using GoThru. GoThru’s BlueLine mode is a mode that is processed using the same logic as the video mode in the Street View app.

Of course, the AI ​​on the Street View server should be trying to understand the top and bottom of the photos you sent first.
However, by determining the top and bottom in the Street View server from the contents of the image, the image in the top and bottom state different from the image uploaded by the user will be published at the time of publication.
The same is true for orientation.
So, as long as the adjacent images are continuous photos, the Street View server AI will automatically correct all the orientations of the photos. I’m grateful for that feature.

For the two tours I introduced in the Top article, I uploaded the same photo to both of them.
For one which used still image mode, Street View was created exactly as the photo I uploaded.
For another one that used the video mode, the Street View server decided the top and bottom of the image. He also “corrected” the coordinates of the image so that it would be consistent with the content of the image, and then drew a blue line.