Train vs test data similarities
From the
QMUL (pixel distance)
QMUL (histo distance)
IIT (pixel distance)
IIT augmented (pixel distance)
HOCoffee (pixel distance)
dataset
, Show distances
from
to
only. (That's gonna be
x
images.)
But only the
0
missclassified ones.
(Oh, and maybe you want to
filter
the testname by the following regexp:
, which matches
0
images.)
Note that the smallest distance is
0
and the largest is
1
.
Confusion matrix
A look at the "predictions"
Test
Train
Distance