Artifacts for the dinoSparseRing example

[drewm1980]

Hello,

I am running the included dinoSparseRing example:

./scripts/dinoSparseRing.sh

and I get the following artifact far from the object:

I observed this behavior even prior to the recently committed changes to the parameters in the example shell scripts; I just hadn't reported it yet.

[yangailin]

I encountered the same problem.

[damonftl]

I'm pretty late to the party here (about three years later) but I'm currently seeing the exact same results on the straight dino dataset. That is, from running /scripts/dino.sh

It's kind of a weird result. It almost looks like a view frustum or something. It's not really random, and it's all black instead of the dino gray color. Not sure what's going on, but I've only managed to get this even running about 30 minutes ago. Hoping to try on my own images and see what happens.

(edit) Actually I just looked at the image set and it's all dino just surrounded by flat black space. That has to be what these points are. Obviously it can't figure out a depth on all-black pixels when every image has them. If you rotate the view just right it looks exactly like the black border around every image, rectangular and everything.

If the author is still watching this repo he should confirm, but I'm 99% sure that's what it is. Any data with real photos of buildings or whatever should not have this problem.

The black points also go away in the final 3D model. At least they did for me with the dino dataset.

[da-carte]

I ran dinoSparseRing and had much better results.

[da-carte]

However, I am now trying to do reconstruction using my own images and it is creating exactly the same artifacts you have been seeing. I don't really understand why.

[damonftl]

I had big problems trying it on my own images too. It wasn't totally just the black points though - the results were really far off. It looked like the points were all randomly dispersed inside a view frustum. I haven't had a chance to figure out what happened yet.

[da-carte]

Yer, so I have taken another set which is clean in VisualSFM with the aim of getting the depth map from Gipuma and it creates horrible output

[damonftl]

Ah yeah that looks very close to what I was getting too, so we seem to be having similar problems. I hope to have some time next week to check this out again. I'll post if I figure anything out.

[da-carte]

I have been able run Gipuma using the cmd line to change starting camera and get a result for each image without the artifacts but I don't know how to fuse them together. Unable to change the bash files to get this to work correctly. I am stuck. There isn't enough information here to get it working properly with the experience I have