modified SIFT to 1) double image before finding keypoints, 2) use floating-point internally instead of 16-bit integers, 3) set the keypoint response to the abs(interpolated_DoG_value). step 1) increases the number of detected keypoints significantly and together with 2) and 3) it improves some detection benchmarks. On the other hand, the stability of the small keypoints is lower, so the rotation and scale invariance tests now struggle a bit. In 2.5 need to make this feature optional and add some more intelligence to the algorithm.
added test that finds a planar object using SIFT.
Fixed visualization by choosing the color appropriate to the detection
score.
Previously the example showed all detections with the same color
disregarding the confidence. This led to the impression that the object
detection did not work at all because there are many detections with low
confidences.
PR to master was
https://github.com/Itseez/opencv/pull/320
- fixed printCudaDeviceInfo for new CC
- fixed some compilation errors and warnings
- removed unset command from CMake script
- removed unused std imports
* added Access parameter to GlBuffer::mapHost
* added autoRelease parameter to all create methods
* fixed indentation in gl_core_3_1
* minor improvments for opengl sample
* removed OpenGLFuncTab, now extensions are loaded internally
* renamed GlTexture -> GlTexture2D
* added support of GlBuffer and GlTexture2D to InputArray/OutputArray
* added ELEMENT_ARRAY_BUFFER and PIXEL_PACK_BUFFER targets
* added copyFrom/copyTo method for GlBuffer and GlTexture2D
* removed GlFont
* removed pointCloudShow
* removed OpenGLCleanCallback
-Change input/output order from (out Labeled, in Image) -> (in Image, Out Labeled) and convert
to Input/OutputArrays in the process.
-Adopt OutputArray for statistics export so that the algorithm is "wrapper friendly" and not requiring a new struct in
language bindings at the expense of using doubles for everything and slowing statistics computation down..
cuda-geek