Hi, I have a new Android phone since my older one broke but this new one seems much more sensitive to the quality of the road surface (bicycle). Now I have about 1 on 20 pictures upside down…
Since I often upload 1000s of images it is too much work to delete or rotate these images.
Question: Is there already a mapillary bot that fixes these kind of upside down pictures automatically without sending a support mail? If not, when it is ready?
So e.g. this one is upside down , Mapillary but the images before are okay ( Mapillary here you see the bad road quality due to the upside down picture) , so it would be not to hard to automate a upsidedown-fixer just looking the location of the (blue) sky…
As a test I rollback to a very old version of the app ( 3.14) on my new phone. With that setup I had never upside down pictures with my old phone…and I have NO upside down pictures at all on the new phone!
So what did you guys add from 3.14 to the newest that makes the camera so sensitive? Please remove that part and I can use the latest version…
This problem became relevant on unpaved road with many vibrations, so probably the app uses the accelerometer to find the orientation for each photo . imho this is a wrong behavior. in many case the device is not rotated after you set it to take the photos and so this detection should be disabled after the first photo orientation is detected.
For the already uploaded photos the process of sending an email with the photo id is tedious and slow…imho in the edit sequence, other than “delete image”, there should be a “rotate image”. better: if your computer vision IA is able, it could detect all rotated images and put them in a check queue in an ad-hoc Verification Project.
yes. So it would be great if the app just uses a fixed ‘good’ orientation based on the first 5 pictures or so made and ignores sudden 180 degrees ‘corrections’
Same here. Using a bike and having quite a lot of uspide down pictures, even at slow speeds… Surprised that we can change picture orientation just like compass angle. Hope it will come…
Hi everyone. Just found this old thread. We have updated the camera, changed big parts of implementation. I hope that this issue will not bother you again. Currently, in the open testing.