ORB_LATCH_K2NN

About

Implementation of komrad36's LATCH and K2NN libraries from komrad36, based on ORB keypoints from OpenCV's features2d

Usage

--index-path <path> Builds a training.dat of all descriptors found in images located at <path> (and a random image is chosen and added to query.dat)

Or to create a predictable set of 255 descriptors of 64 bytes each (512-bit) into training.dat

--test-index Builds a training.dat of all descriptors found in images located at <path> (and a random image is chosen and added to query.dat)

Last byte of each 8 byte descriptor is set 0..255

 [first 7 byte]..[last byte]
 11111111111111..0000001
 11111111111111..0000010
 11111111111111..0000011
 ...
 11111111111111..1111101
 11111111111111..1111110
 11111111111111..1111111

--search Search all of training.dat for the nearest neighbours (K2NN) from descriptors in query.dat

Goal

To be able to find images sorted by their "similarity" in terms of ORB feature similarity to detect tshirts with similar designs on them but taken by different photographers at different distances.

Problems

None of the descriptors are matche at all

./ORB_LATCH_K2NN --search
Loaded 64 bytes from query.dat
Loaded 16320 bytes from training.dat
First descriptor in query.dat looks like.. 
11111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111100000000
First descriptor in training.dat looks like.. 
11111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111100000000
---- fastApproxMatch results ----
Warming up...
.....
No matches found
---- bruteMatch results ----
Warming up...
.....
No matches found
---- exactMatch results ----
Warming up...
.....
No matches found

Todo

Use CUDA libraries once basics are working