Photogrammetry environment assets creation learning

74m4

Hello everyone. I want to learn the process of creation different environment assets for realtime PBR with the help of photogrammetry.
For now i have some basic understanding of the whole process. So the this topic is some kind of journal. and I hope some advices. Thx.
For now the pipe line is some kind of:

1. Make some photos. 
2. Create hipoly mesh with the help of photogrammetry software tools
3. Create low poly mesh (aproximately 2k triangles)
4. Bake normal, color maps to low poly mesh
5. Some magic with maps with the help of substance

Yesterday night i made some test.
1. Created several photos (approximately 40) of vase (LG Nexus 5x phone standart camera)

Add markers:

to be continued... have to go.

74m4

Align cameras:

Generate dense cloud, and delete unwanted vertices:

After that generate Model and Texture:

I have got a nice looking vase, with some strange neck due to very bad light and upside down position. But i'm very glad with the result. 
sketchfab link https://sketchfab.com/3d-models/vase-boom-112124bf1d91460d8221948ec86d6de2

zachagreg

You need to get a greater range of angles in order to capture the shape accurately. Currently all of your photos are only from the top 45 degrees looking down. 

JohannesAg

As already pointed out you need way more angles. Mike pavlovich made a small intro to photogrammetry series on his youtube channel that goes over this stuff pretty well, It´s worth taking a look at.

74m4

zachagreg said:

You need to get a greater range of angles in order to capture the shape accurately. Currently all of your photos are only from the top 45 degrees looking down. 

JohannesAg said:

As already pointed out you need way more angles. Mike pavlovich made a small intro to photogrammetry series on his youtube channel that goes over this stuff pretty well, It´s worth taking a look at.

thx a lot )

74m4

Meanwhile have found nice stump. made several photos. 

74m4

Align cameras.

Generate low quality mesh and made masks for photos from that mesh.

Due to bad quality of JPG photos, and lack of experience, has to generate medium quality mesh. Not enough data in photos for dense cloud so with ultra high quality there are a lot of holes. So i experimented with different building quality and stopped with the following result.

zachagreg

Doesn't the 5X have the ability to shoot in DNG as well? Worth a shot to see if it can either through lightroom on the phone or through Google camera app. If either of those are available to you.

tinyworlds

Not for the stump, but for smaller, movable objects it helps to place it on or next to something with an unique pattern. E.g. a newspaper. This gives Agisoft a good source to recognize patterns, which it uses to calculate how the images overlap.

Here are 2 great resources to get better results:
https://sketchfab.com/blogs/community/how-to-set-up-a-successful-photogrammetry-project/
http://www.agisoft.ru/w/index.php?title=PhotoScan/Tips_and_Tricks