Using bicycle dashcam video, no usable results

Thats only the tail of the output, since it ram for so long and printed a lot I guess.

What are you trying to generate?

Ideally I'd like to have orthophotos to use for OpenStreetMap

You won’t get orthophotos from a camera facing forwards. You need nadir images, or at least images angled towards the ground…

See https://docs.webodm.org/flying-tips/ for recommendations

You actually should be able to get orthophotos from street imagery (e.g. https://jakecoppinger.com/2022/12/creating-aerial-imagery-with-a-bike-helmet-camera-and-opendronemap/) here it just looks like reconstruction failed for some reason

Share the dataset if you can 🙏

Oh is it a 360 camera, or just forward facing? Maybe I misunderstood.

Hero8 is not 360 so I wouldn’t have thought it would capture the required detail?

No, the hero8 I used is not 360.
I would have thought that so much front-facing data would give me at least some kind of results though.
But I guess if the view-angle of the setting I used isn't big enough?
It's a bit disappointing though.

That's an example of the kind of images.
I don't want to make it public as it contains other license plates, people, faces and I don't want to anonymize it just for that

As I understand it, you kind of need it to capture images from all angles (or at least, multiple angles) so that it can build up the 3D model, and for it to build an orthophoto (ie 2D top-down) it would need to be above the features so that it’s able to capture the features and match the images to build the ortho. 🤔

Hm, so it'd be better to have gone the same path forward and backward?
Essentially simulating an array of cameras via multiple passes instead of using multiple cameras at once.

Still, I'd assume that driving past something means it has enough angles of at least some of the things, mostly of that which is close to the kerb.

I think you’d at least have better chance of it finding matching features if you did that 🤷♂️

right, of course, ok.

https://www.mapillary.com/app/?pKey=1633345693801133
I tried to use a few 360 images, downloading my sequence from mapillary, and my pc now runs out of memory during the renderdem step.
I have 32G of RAM, and (for something else I don't remember) I have 80G swap, both were full.
I set feature-quality and pc-quality to lowest, I don't see any other obvious setting which would influence memory usage, I hope that's enough.
Or I could use less Images, but then I'd need to stitch final outputs. which I'd like to avoid

I don't know what he did so differently, but my attempts at this have failed so far.
enabling bg-removal and sky removal helped though.
but I don't get recognizable point clouds.
with other settings (camera-lens: equirectangular, as it's what exiftool says the projection type if) I only got an error; "This is likely an unsupported camera model."

Jake Coppinger also mentions "You will likely need to enable “resize images” to 2000px or it is very easy to run out of memory."; I don't see that option though.

Did you try setting the camera to spherical?

Yes