대규모 프로젝트 워크플로우 알려주세요

International 한국어 (Korean)
, ,
1

저는 약 800gb 이미지를 사용하여 대규모 프로젝트를 완성하고자 합니다.

*컴퓨터 사양 RTX4090/i9-13900K/RAM64 * rcv:1.2.2

  1. 정렬

우선 모든 이미지를 불러와서 한번에 정렬을 시도하고 있습니다.

그런데 너무 오래 걸립니다. 1주 넘어도 완성을 안 합니다.

그래서 정렬 옵션을 최하로 선택하고 일단 진행 하고 있습니다.

(Q:적절한 퀄리티 있는 옵션을 선택하여 고용량 정렬은 완성하는 방법을 알려주세요)

  1. CIL 사용하여 한번에 FBX 파일을 내보내는 작업을 시도 하고 있습니다.

그런데 중간에 캐시가 쌓여서 프로젝트가 멈춰 있습니다.

(Q:CIL 스크립트 중에 자동 저장 및 캐시제거 옵션을 알려주세요)

  1. 노멀디테일 시간이 너무 오래 걸립니다. 한 구역 당 10~24시간 걸립니다.

테스트로 모델링 구경도 못하고 있습니다.

만약 정렬 퀄리티가 낮다면 모든걸 다시 해야합니다.

(Q:속도 빠른 노멀 세팅 알려주세요)

  1. 혹시 제가 모르는 대규모 프로젝트 진행 하는 방법이 있다면 알려주세요

리얼리티켑쳐에서 알려주는 모든 대규모 프로젝트를 보고 있는데

따라할 수가 없습니다…

2

Hi realroadstudio,
how many images is those 800GB? What are your alignment settings?
The pre-defined alignment settings are the best for most datasets. What kind of images are you using?
It is not a good idea to clear a cache during computation, but you can do it after some steps (alignment, meshing). But it is better to keep them until your model is finished. But with so big images you will need terabytes for your cache.
What are your reconstruction settings?
For such big models you can use 3D divider script: Processing of the Habitat 67 Scan Data with the 3D Divider Script | Tutorial

3

  1. 8k / 59,763 image

  2. Max features per image : 5000

  3. he pre-defined alignment settings are the best for most datasets. (ok)

  4. dron image / 5 direction

  5. It is not a good idea to clear a cache during computation, but you can do it after some steps (alignment, meshing). But it is better to keep them until your model is finished. But with so big images you will need terabytes for your cache.

So how did you manage to take on such a big project? I have a 2TB SSD

  1. What are your reconstruction settings? Basic setup

  2. From alignment to modeling, tell us procedurally how to proceed with a large-scale project. Also, please let me know if there is a way to do it with multiple computers.

4

According to the number of images, your alignment settings and RAM the alignment should be done.
Memory consumption during the alignment phase depends on the number of images (not size) and the number of detected features per image. For the default setting of 40 000 features per image (Alignment setting), you can expect the following boundaries:

  • 2,000 images - 16GB RAM
  • 4,000 images - 32GB RAM
  • 8,000 images - 64GB RAM
  • 16,000 images - 128GB RAM
    By decreasing the number of detected features to half you can approximately decrease the memory consumption by half as well. The approximate formula is: RAM = features x images x 200 bytes.

As you wrote 5 direction for images, what do you mean by that? Is it possible for you to align the separated directions and then merge those components together?
Is this 2 TB SSD just for the cache? A medium-sized city project can easily go to 25+TB for 1 single project (inputs, iterations, models, textures, etc.).

For your project I would align each camera direction separately and export components. Then I would align those 5 components to one component. Then I would use divider script which I sent you before, divide the reconstruction region to a smaller parts and process those parts as models separately. For this you can use also more computers (you just need to copy the basic project with alignment, created reconstruction regions and process the models). But you need to have license on each computer where you want to use RealityCapture.