1.0.5
Fixed a bug (rarely happens) in CUDA kernel configuration
	* Symptom: ComputeJtE<Point>: invalid configuration argument(9)
	  Triggers when npoint is within (2 * n * 65505 ~ 2 * n * 65535) for n >= 2
Changed the CUDA error handling
	* from exit(0) to throw "error location string"
Added Support for constraints of camera clusters having equal focal lengths
	* Call ParallelBA::SetFocalMask(reference_camera_index_for_each_camera, weight);
	* For example, a vector of 0 means all camera should have the same focal length.
	* For example, <0,0,2,2,2> gives two groups (0,1) and (2, 3,4) for the constraint.
Improved memory allocation
	* Reset device and deallocate all data when ReserveStorage(0, 0, 0);

1.0.4
Added AVX-based implementation (Use Release_AVX for MSVC, or #define CPUPBA_USE_AVX)
Added function ParallelBA_GetVersion for future changes.
Fixed minor bug in the profiling program

1.0.3
Added support of constant cameras (CameraT::SetConstantCamera)
	* Causes slight slow-down of standard pipeline 
	* Can be disabled by defining PBA_DISABLE_CONST_CAMERA
Automatic switching from GPU to CPU if CUDA is not supported
	* Will not switch back 
Fixed a problem with Windows DLL function name
	
1.0.2
Choosing thread count accoding to the number of CPU cores. 
	* This might be slightly better than the fixed parameters
	* This is still very heruristic and not really optimal.
	* The settings are still tunned for very large data. 
Added two additional Bundle Adjustment Modes (Motion only & Structure only)
	* Call ParallelBA::SetNextBundleMode() to set the mode

1.0.1
Converted MSVC solution files to visual studio 2010
Added the first draft of manual
Added CMakeFile (Provided by Pierre Moulon)

1.0.0 (06/2011)
The first release