We just started the project here about a week ago. We are using a go pro max2 360 video at 4k probably driving around 20-30mph. When we upload the video and it converts them to images. The images look grainy and our street signs aren’t being identified on top of our poles. I see some posts about switching to photos. Would this be the recommendation over video for higher quality?
Thanks for your report. Mapillary works with images, not videos; that’s why the video is sliced and converted to images. We don’t keep videos on our side. At the same time, images shouldn’t be grainy. If you share your captures (e.g. Mapillary link), we can verify why road signs are not being identified and whether the images look reasonable (or are too grainy).
__
BR, Yaro
Any help would be appreciated. I’m just trying to figure out if we pump it up to 8k or go for still images if the 4k isnt grabbing a high enough resolution for the AI to differentiate.
I get the impression the detection is working for your example;
It does however appear that the camera focus is being affected by the vehicle body. ie compare the inside tailgate with distant trees. Could also be vibration?
I randomly selected a handful of sequences for a check, and it looks like the detections are there. Maybe the issue was the processing and propagation delay, as it takes 24h at best and up to 72h (sometimes more) for data to be propagated. Alternatively, please share an image where the detection didn’t work.
Regarding 4K vs 8K, I would say 8K is better. The images are 360, and 8K is the default for Max 2, so I would suggest using it if possible.
__
BR, Yaro