Raw data exploration completed, and can be found in this notebook.
Main comments:
Label classes are highly imbalanced: Number of objects by types range from 325 (ground-track-field) to 28,068 (ship)
Co-occurences of object types are not surprising. High correlations are between ground-track-field/soccer-ball-field and helicopter/plane, because these tuples are tend to be in the same images. Again uninterestingly, planes and tennis-courts are negatively correlated, since it is unlikely to have a tennis court near an airport.
Do an explorative analysis on jupyter notebook and put it to /notebooks/explore-raw-data
Notebook should iterate followings:
Explore raw training data from DOTA, and report followings:
Show some image samples
Write an overall summary of explorative analysis. And add necessary information from DOTA paper in it.