According to DSO paper for Step 2: Keyframe Creation we need to calculate simply and straightforwardly 2 mean square optical flows f and f_t.
As I understand, they are calculated in CoarseTracker.cpp - here and these calculations are strange and discouraging for me - I can't understand why we need additional residual with negative (according to code comments) transformation ((R, -T) for f and (I, -T) for f_t) for every usual residual with positive transformation ((R, T) for f and (I, T) for f_t).
I can't understand why only 1 positive transformation isn't enough, why we need so special additional negative transformation with negative translation only and what it gives to us - why it is better.
Hello!
According to DSO paper for
Step 2: Keyframe Creation
we need to calculate simply and straightforwardly 2 mean square optical flowsf
andf_t
.As I understand, they are calculated in CoarseTracker.cpp - here and these calculations are strange and discouraging for me - I can't understand why we need additional residual with
negative
(according to code comments) transformation ((R, -T)
forf
and(I, -T)
forf_t
) for every usual residual withpositive
transformation ((R, T)
forf
and(I, T)
forf_t
).I can't understand why only 1
positive
transformation isn't enough, why we need so special additionalnegative
transformation with negative translation only and what it gives to us - why it is better.Can anyone help me please? Thanks in advance!