Closed qxcv closed 8 years ago
See train/random_stitching_weights.m
and train/stitching_weight_search
. It seems to work, but for the fact that it's really slow.
Not sure what I should actually run it on, though. Maybe the validation set?
This basically works, but I'll leave the issue open until I've gotten a final set of weights for each dataset.
Closing this, since I've already got some sensible weights.
Some sort of grid search (or maybe randomised search, which is meant to get better results in much less time) would really help figure out what the best stitching weights are.