pull request

[edgarriba]

@dakotabenjamin @smathermather @paulinus

The baby starts to walk. Let's test everything in order to do a proper PR in the master branch.

We can also discuss here the project design in order to be able to move things around before doing this huge merge.

[dakotabenjamin]

What kind of parallelization is happening, specifically with opensfm? These processes seem to be taking longer than normal, even on a large (32-core) machine. I'm trying to do some comprehensive testing but right now run time is slowing me down.

It took 45 minutes to run odm_data_bellus, vs 18 minutes using the master

[edgarriba]

AFAIK parallelization happens only during the features detection and matching process. Not sure if Ceres has parallel computing for BA task? @paulinus

[dakotabenjamin]

Can you help me get the python testing started? My unfamiliarity with ecto really has me struggling to set up the tests. I think I understand conceptually that we need to initialize an ecto plasm, then we can call each cell and run tests on each- however I am having trouble getting that basis. You can see here what I mean: https://github.com/dakotabenjamin/OpenDroneMap/blob/nose_tests/tests/test_odm.py

[edgarriba]

sure! Firstly, in this case what do you want to test? Secondly, what I would do is take one of the datasets that produces a proper result and use it as a ground truth reference. Finally, design a test for each cell using as inputs/outputs the ground truth files. We should define very well every test before start coding.

[dakotabenjamin]

Ok So I think this is a start of what should be tested in python. I will use a subset of the toledo dataset.

opendm/

config.py

• test each parameter (min, max, out of bounds, etc.)

context.py

• Each path must be a valid, existing path

io.py

• check each function with a known string

system.py

• (Get_ccd_widths is depreciated)
• parse_coordinate_system:
  • This is also old code - new one is in types

types.py

• ODM_Photos
  • updated focal equal to known focal length
  • updated ccd equal to known ccd
• ODM_Georef
  • calculate_epsg needs to be updated, see master
  • convert_to_las: assert existence of las file output
  • utm_to_latlon:

scripts/

dataset.py

• test that supported_extensions works with a variety of file names, even bogus ones
• check outputs are all the photos

resize.py

• resulting images are the right size
  • img.shape[0] or img.shape[1] == params.resize_to
• metadata has been properly updated
  • img.metadata['Exif.Photo.PixelX/YDimension'] == params.resize_to

OpenSfM.py

• config file is contains info same as params
• when matcher_distance > 0 it is written to the config
• at least one reconstruction file is generated
• check that bundler file is exported (possible to check if valid?)

CMVS.py

• validate params
• system.run() command is equal to some known string

PMVS.py

• validate params
• system.run() command is equal to some known string

odm_meshing.py

• validate params
• system.run() command is equal to some known string

odm_texturing.py

• validate params
• system.run() command is equal to some known string

odm_georeferencing.py

• validate params
• system.run() command is equal to some known string when using EXIF coords
• system.run() when using GCP

odm_orthophoto.py

• validate params
• system.run() command is equal to some known string

[edgarriba]

sounds good to me! We could use this unit test framework: https://cloud.google.com/appengine/docs/python/tools/localunittesting?csw=1

[dakotabenjamin]

Interesting. I had been looking at using Travis-CI because of the strong support for open-source projects, and it integrates with Github well. What do you think google app engine can offer that would better suite this project?

[edgarriba]

well, I'm not confident at all with Travis since I've never used. But we can give it a shot! Regarding gtest I have some experience with the c++ framework (used by opencv) but not in python. I only know that have some well defined macros for testing.

[dakotabenjamin]

I finally got a testsing framework in place. I made a branch on the main repo from the cmake branch for the code sprint and I've pushed the testing code to as 'tests' folder there.

BTW, I had to use the git merge option -s recursive -Xtheirs to basically completely override your code into the branch.

[edgarriba]

yes, just seen that. Thanks to merge it. I realized that the sprint starts tomorrow. I'm a bit busy but I will help as much as possible.

[smathermather]

Sounds great, @edgarriba.