search

neptune-ai / open-solution-mapping-challenge

Open solution to the Mapping Challenge :earth_americas:
https://www.crowdai.org/challenges/mapping-challenge
MIT License
378 stars 96 forks source link

Blank masks after preparation step #172

Closed maksimovkonstantin closed 5 years ago

maksimovkonstantin commented 5 years ago

@jakubczakon Hi!I try to use your great work with my local python env, but all masks that produced on preparation step is blank (png images) or it is ok that they are blank?

jakubczakon commented 5 years ago

Hi @maksimovkonstantin .

It should definitely not return blank masks. Potentially the paths are specified incorrectly in the neptune.yaml . Would you mind pasting a snippet with those paths here (or a mock that shows the picture)?

maksimovkonstantin commented 5 years ago

@jakubczakon Another problem that the preparation code was stucked for 2 days(((( using native python

jakubczakon commented 5 years ago

That is really long. It was under one day for us though with those parameters:

  num_workers: 4
  num_threads: 1000

changed to something higher than default values.

maksimovkonstantin commented 5 years ago

@jakubczakon if I have 12 threads and 128 gb ram - it will be ok to setup num workers to 10 and num_threads to let's say 3000?

jakubczakon commented 5 years ago

Yeah, I think so.

apyskir commented 5 years ago

@maksimovkonstantin This is no surprise that these .png files look like being empty, but probably they're not. When I display my own png files they seem to be black, but I know, they're not empty. Try to open them with PIL.Image.Open() and check if underlying numpy.array is empty. Good luck!

kamil-kaczmarek commented 5 years ago

Seems like we have covered this question.