reporting_notes.txt

# OpenSfM report design


Two concepts:
 - logging: write down stuff you are doing as you do them
 - reporting: write a summary of the execution statistics


Logging is always done on-the-fly. Text or json are logged as processes do stuff.

For reporting we need can choose to build the report on the fly or as a post step.  Doing it on-the-fly is more efficient since there is no need to reload data from disk.  Doing it as a post process has many advantages: it is simpler to implement, specially for multiprocessing. It is decoupled from the main code. And enables to build more global measures.

Some measures, like running time, need to be computed on-the-fly.

Possible solution:
- each process or function can write small independent reports as independent files.
- a post process gathers all the information and builds a full report.


- dataset
    - reports
        - features
            - counts
                - image00.jpg.json
                - image01.jpg.json
                - ...
            - report.json
        - matches
            - counts
                - image00.jpg.image01.jpg.json
                - ...
            - report.json
        - tracks
            - report.json
        - reconstruction
            - report.json