A4A1 - harris parameters

Moderator: Computer Vision

banshee
Nerd
Nerd
Beiträge: 684
Registriert: 22. Okt 2006 18:46

A4A1 - harris parameters

Beitrag von banshee »

hey,

I'm wondering if we should really use a threshold of 1500 because if I do so, I won't get any interest points because the maximum harris function value for all images seems to be 196 or am I using the harris implementation incorrectly? I just input my images as a 200x<sth> uint8 grayscale images...

/e: Alright I found it. All input images have to consist of doubles (or another data type of enough space) because if it was uint8 the lsum and lprod variables are capped at 255 resulting in way too small (and wrong) harris values.

robert.n
Nerd
Nerd
Beiträge: 673
Registriert: 29. Sep 2008 19:17

Re: A4A1 - harris parameters

Beitrag von robert.n »

banshee hat geschrieben:/e: Alright I found it. All input images have to consist of doubles (or another data type of enough space) because if it was uint8 the lsum and lprod variables are capped at 255 resulting in way too small (and wrong) harris values.
Thanks, I ran into the same problem.

What do your test and training errors look like? Apparently my classifier makes no mistakes at all. Is that possible?

tanne
Endlosschleifenbastler
Endlosschleifenbastler
Beiträge: 162
Registriert: 30. Sep 2008 16:05

Re: A4A1 - harris parameters

Beitrag von tanne »

robert.n hat geschrieben:
banshee hat geschrieben:/e: Alright I found it. All input images have to consist of doubles (or another data type of enough space) because if it was uint8 the lsum and lprod variables are capped at 255 resulting in way too small (and wrong) harris values.
Thanks, I ran into the same problem.

What do your test and training errors look like? Apparently my classifier makes no mistakes at all. Is that possible?
sounds pretty unrealistic. ^^
mine works quite well, except with the motorbike test datas..i have about 40% error there..but only there. dont know why. :/

robert.n
Nerd
Nerd
Beiträge: 673
Registriert: 29. Sep 2008 19:17

Re: A4A1 - harris parameters

Beitrag von robert.n »

What are the definitions of testing and training error?

Because I haven't found a formula in the slides I am currently just counting the number of correct and incorrect classifications.
But of course the confidence scores are not always [1, 0] or [0, 1]. Sometimes they're even like [0.33, 0.67] etc.

qgao
Moderator
Moderator
Beiträge: 33
Registriert: 7. Apr 2009 11:38

Re: A4A1 - harris parameters

Beitrag von qgao »

robert.n hat geschrieben:What are the definitions of testing and training error?
training error: rate of correct recognition w.r.t. training data;
testing error: rate of correct recognition w.r.t. testing data.

Antworten

Zurück zu „Computer Vision“