Why CPU version generates much more keypoints than GPU one?

[Dmitry Klimenkov]

Here is the result ( 4671 VS 316.68)

### FAST detector
 CPU ---
 FAST: min: 4559, max: 6822, avg: 5477.3 [usec]
 FAST feature count: min: 3258, max: 7875, avg: 4671.76 [1]
 GPU ---
 FAST: min: 677, max: 971, avg: 716.67 [usec]
 FAST feature count: min: 270, max: 340, avg: 316.68 [1]
 Note: No verification performed
 Success: OK

Any ideas?

[Dmitry Klimenkov]

Ah, just read the point in the paper about uniform distribution of the features, so lowering cell size increases the number of detected features