search

OpenDroneMap / ODM

A command line toolkit to generate maps, point clouds, 3D models and DEMs from drone, balloon or kite images. 📷
https://opendronemap.org
GNU Affero General Public License v3.0
4.88k stars 1.11k forks source link

Memory issues with large dataset #856

Closed nj07 closed 6 years ago

nj07 commented 6 years ago

I installed OpenDroneMap using Docker (Docker version 17.12.0-ce) and running it on Ubuntu. Operating System: Ubuntu 16.04 LTS OSType: linux Architecture: x86_64 CPUs: 12 Total Memory: 23.46GiB

My input size is 375 images. I have tried running OpenDroneMap with and without any additional options. I have also tried it with passing option for 1 processor.

docker run -it --rm -v "$(pwd)/images:/code/images" -v "$(pwd)/odm_prthophoto:/code/odm_orthophoto" -v "$(pwd)/odm_texturing:/code/odm_texturing" opendronemap/opendronemap --opensfm-processes 1 --fast-orthophoto

But it always fails:

Load and prepare mesh: PLY Loader: comment VTK generated PLY File Reading PLY: 100075 verts... 200001 faces... done. Generating texture views: NVM: Loading file... NVM: Number of views: 374 NVM: Number of features: 0 Loading 100%... done. (Took 79.995s)
Building adjacency graph: Adding edges 100%... done. (Took 0.514s)
299928 total edges. View selection: Building BVH from 200001 faces... done. (Took: 204 ms) Calculating face qualities 100%... done. (Took 53.958s)
Postprocessing face infos 100%... done. (Took 0.235s)
Maximum quality of a face within an image: 384729 Clamping qualities to 11696.9 within normalization. Writing data cost file... done. Optimizing: Time[s] Energy 0 203079 1 201308 1 198908 1 197641 1 197033 1 196675 2 196482 2 196349 2 196210 2 196129 3 196076 3 196039 3 196004 3 195968 32734 faces have not been seen Took: 58.531s Generating texture patches: Running... Killed Traceback (most recent call last): File "/code/run.py", line 47, in plasm.execute(niter=1) File "/code/scripts/mvstex.py", line 139, in process '{keepUnseenFaces}'.format(**kwargs)) File "/code/opendm/system.py", line 34, in run raise Exception("Child returned {}".format(retcode)) Exception: Child returned 137

It is always running into memory issues. What is the minimum memory required to process ~400 images. Is there a benchmark? Please help. Thanks.

pierotofy commented 6 years ago

The codebase is changing frequently, so we don't have definite benchmarks as of yet (but we have plans to create one in the future). It's not just a function of number of images, but also of resolution. I'd recommend scaling down your images to go past the texturing step (or use WebODM, which does it for you).

kikislater commented 6 years ago

Does resizing influence focal lens and sensor size ?

smathermather commented 6 years ago

@kikislater, No need to worry about changes to focal lens nor sensor size. These are independent of image pixel dimensions.

kikislater commented 6 years ago

Thank you Stephen

nj07 commented 6 years ago

@pierotofy, I keep running frequently into memory issues or Saving model... libpng error: Write Error, opendronemap. I have large datasets to process (> 500 images). I am resizing the images to 1200. I am running Google Cloud and have allocated 100Gb of memory and 50Gb of hard disk. Can you recommend anything else that I could do? Getting a failure after almost 8 hours of processing is kinda disheartening. On another note, opendronemap doesn't seem to be GPU enabled. Is that right?

pierotofy commented 6 years ago

A write error is likely due to not enough disk space... add some more disk space.

You are correct, we currently don't have support for GPUs.