I'm writing some photoshop scripts to process a bunch of images before they get uploaded as imagetargets...so I needed to figure out what process to put all of the source images through. The first question was, "what size should I save them out to?" I kind of guessed that it didn't matter since vuforia is scanning for relative features...so my source image needs to be "big enough" to capture the features, but making it bigger wouldn't give me an improvement. That was my assumption. So I tested this by uploading the same image at various sizes.
The original size happened to be 830
I doubled it
Halved it
Also tried some that were neatly in powers of 2 (old 3d graphics assumptions)
One below the recommended size
Everything scored a 4/5 on the augmentable rating except for the original and doubled size which scored a 3/5. I had expected the original to be the highest since it contains the least artifacting, but I was wrong. When I view the identified features of the various test sizes (using the target manager), they don't really look different. Maybe they have more or less the same number of features, but the rating is determined by taking a ratio of features to pixels or something.
Also - I imagine that the augmentable rating has a bunch of in-betweens on the ratings, but for sake of being simple we just see a 1 - 5. In reality maybe one of my 4.0 targets is better than the rest of the 4.0 targets.
I guess my question is...how do I figure out what the ideal upload size is?
good for imagetargets?
There is no ideal upload size, in general.
in your case, it could simply be that your rating is technically very close to 3.5 in all cases, which is then rounded to either 3 or 4 stars.
Have you tried the 3 stars and the 4 stars at runtime using ImageTargets sample for instance ?
good for imagetargets?
No I don't see much of a difference, but it's pretty subjective for me to say one way or another. I was actually thinking of writing a tool that showed me practical performance on the device itself.