Forum

Author Topic: Depth map generations works only partially - solutions  (Read 1082 times)

MikeZ

  • Newbie
  • *
  • Posts: 23
    • View Profile
Depth map generations works only partially - solutions
« on: April 28, 2019, 04:53:19 PM »
Hi guys. At the beginning, I would like to say that Agisoft is probably my favorite software. I like the way you develop it and also how responsive makers are. I've read so many posts by Alexey that I feel like we are close friends :). I use agisoft for a few years. In the past, I had a chance to use pro license now I use my private standard. I've reached the point when reading only is not enough anymore. I need to ask some questions. Reason for that is one. I'm planning a huge digitization project and I need to choose the right software for it. It will be super cool if Agisoft can do it but I need to be sure.

I'm gonna write separate posts with my questions. It always produces nice content.

As it was explained before: Agisoft won't generate a depth map for cameras that positions are in the field of view of another camera. That was explained on "the corridor" example. Everything is fine but:

I have a test scan. It contains 4k cameras. I choose one part to realign. I chose photos that were used to reconstruct a small sculpture. It was 95 cameras. Reconstruction was done quite nicely. So I thought if (as input was each 15th frame from footage)i will get a way better result from whole footage. I imported around 1700 frames of the subject. I aligned it properly and... I was able to generate only 150 depth maps. What is worst those depth maps were not corresponding to those that were generated at first attempt. I've lost half of the statue. After reading forum i decided to turn off half of the cameras. The effect was worst, 90 depth maps. I disable manually some cameras that can cause problems. Nothing happened. I have disabled 90% of cameras so i had the similar representation of frames as at first attempt. from 100 cameras i received only two depth maps.

How is it possible that from one data set, a similar quantity of cameras I received quite opposite results? How I can control this issue. How i can select and erase causing problems cameras?

The footage was made by smartphone in extremely bad light conditions. A lot of frames are blurry and unsharp. As I said before, this is an exercise that is aiming to prepare workflow for big project. In this scenario, there is no chance to capture photos again.

cheers guys


Edit:
Another realigment of the statue and from 1758 frames i have 1737 aligned. 884 depth maps. It seems to be quite mysterious. I have not disabled any camera.

It seems that it depends on alignment. When i use pair selection i have the worst result. It was mentioned before that it can depends on the number of matches between the photos. Is there any line that we can draw that describes the minimum? Do tweak with maximum neighbors can influence depth map generation?

Edit 2
It's definitely about the maches ... camera optimalisation erased 800 fro 880 depth maps
« Last Edit: April 28, 2019, 05:33:51 PM by MikeZ »