Autonomous Space Robotics Lab Speeded Up SURF fasthessian.cu Go to the documentation of this file. /* Copyright (c) 2010, Paul Furgale and Chi Hay Tong All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. * The names of its contributors may not be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "fasthessian.h" #include "gpu_utils.h" #include "gpu_globals.h" namespace asrl{ __device__ float evalDyyByArea(float x, float y, float t, float mask_width, float mask_height, float fscale) { float Dyy = 0.f; Dyy += iiAreaLookupCDHalfWH( x , y, mask_width, mask_height); Dyy -= t * iiAreaLookupCDHalfWH( x , y, mask_width, fscale); Dyy *= 1/(fscale*fscale); return Dyy; } __device__ float evalDxxByArea(float x, float y, float t, float mask_width, float mask_height, float fscale) { float Dxx = 0.f; Dxx += iiAreaLookupCDHalfWH( x , y, mask_height, mask_width); Dxx -= t * iiAreaLookupCDHalfWH( x , y, fscale , mask_width); Dxx *= 1/(fscale*fscale); return Dxx; } __device__ float evalDxy(float x, float y, float fscale, int octave) { float center_offset = d_octave_params[octave].dxy_center_offset * fscale; float half_width = d_octave_params[octave].dxy_half_width * fscale; float Dxy = 0.f; Dxy += iiAreaLookupCDHalfWH(x - center_offset, y - center_offset, half_width, half_width); Dxy -= iiAreaLookupCDHalfWH(x - center_offset, y + center_offset, half_width, half_width); Dxy += iiAreaLookupCDHalfWH(x + center_offset, y + center_offset, half_width, half_width); Dxy -= iiAreaLookupCDHalfWH(x + center_offset, y - center_offset, half_width, half_width); Dxy *= 1/(fscale*fscale); return Dxy; } __global__ void fasthessian_kernel(float * d_hessian, int octave) { // Determine the indices in the Hessian buffer int hidx_x = threadIdx.x + __mul24(blockIdx.x,blockDim.x); int hidx_y = threadIdx.y + __mul24(blockIdx.y,blockDim.y); int hidx_z = threadIdx.z; // int hidx_lin = hidx_x + d_octave_params[octave].x_size * hidx_y + d_octave_params[octave].x_size * d_octave_params[octave].y_size * hidx_z; int hidx_lin = hidx_x + d_hessian_stride[0] * hidx_y + d_hessian_stride[0] * d_octave_params[octave].y_size * hidx_z; // Get the mask height __shared__ float t; if(threadIdx.x == 0 && threadIdx.y == 0 && threadIdx.z == 0) { t = d_octave_params[octave].mask_height; } __syncthreads(); // Compute the scale float fscale = d_hessian_scale[ASRL_SURF_MAX_INTERVALS*octave + hidx_z]; // Compute the lookup location of the mask center float x = (hidx_x * d_octave_params[octave].step + d_octave_params[octave].border); float y = (hidx_y * d_octave_params[octave].step + d_octave_params[octave].border); // Scale the mask dimensions according to the scale if(hidx_x < d_octave_params[octave].x_size && hidx_y < d_octave_params[octave].y_size && hidx_z < d_octave_params[octave].nIntervals) { float mask_width = d_octave_params[octave].mask_width * fscale; float mask_height = d_octave_params[octave].mask_height * fscale; // Compute the filter responses float Dyy = evalDyyByArea(x, y, t, mask_width, mask_height, fscale); float Dxx = evalDxxByArea(x, y, t, mask_width, mask_height, fscale); float Dxy = evalDxy(x, y, fscale, octave); // Combine the responses and store the Laplacian sign float result = ( Dxx * Dyy ) - d_octave_params[octave].dxy_scale*(Dxy*Dxy); if(Dxx+Dyy > 0.f) setLastBit(result); else clearLastBit(result); d_hessian[hidx_lin] = result; } } __global__ void eval_component_kernel(float * d_hessian, int octave, int component) { int hidx_x = threadIdx.x + blockIdx.x * blockDim.x; int hidx_y = threadIdx.y + blockIdx.y * blockDim.y; int hidx_z = threadIdx.z; int hidx_lin = hidx_x + d_hessian_stride[0] * hidx_y + d_hessian_stride[0] * d_octave_params[octave].y_size * hidx_z; __shared__ float t; if(threadIdx.x == 0 && threadIdx.y == 0 && threadIdx.z == 0) { t = d_octave_params[octave].mask_height; } __syncthreads(); float fscale = d_hessian_scale[ASRL_SURF_MAX_INTERVALS*octave + hidx_z]; float x = (hidx_x * d_octave_params[octave].step + d_octave_params[octave].border); float y = (hidx_y * d_octave_params[octave].step + d_octave_params[octave].border); if(hidx_x < d_octave_params[octave].x_size && hidx_y < d_octave_params[octave].y_size && hidx_z < d_octave_params[octave].nIntervals) { float mask_width = d_octave_params[octave].mask_width * fscale; float mask_height = d_octave_params[octave].mask_height * fscale; float Dyy = evalDyyByArea(x, y, t, mask_width, mask_height, fscale); float Dxx = evalDxxByArea(x, y, t, mask_width, mask_height, fscale); float Dxy = evalDxy(x, y, fscale, octave); float censure = 0.f; float result = ( Dxx * Dyy ) - d_octave_params[octave].dxy_scale*(Dxy*Dxy); switch(component) { case FH_DXX: d_hessian[hidx_lin] = Dxx; break; case FH_DYY: d_hessian[hidx_lin] = Dyy; break; case FH_DXX_DYY: d_hessian[hidx_lin] = Dxx * Dyy; break; case FH_DXY: d_hessian[hidx_lin] = Dxy; break; case FH_LINEAR_IDX: d_hessian[hidx_lin] = hidx_lin; break; case FH_HIDX_X_IDX: d_hessian[hidx_lin] = hidx_x; break; case FH_HIDX_Y_IDX: d_hessian[hidx_lin] = hidx_y; break; case FH_FSCALE: d_hessian[hidx_lin] = fscale; break; case FH_X: d_hessian[hidx_lin] = x; break; case FH_Y: d_hessian[hidx_lin] = y; break; case FH_RESULT: d_hessian[hidx_lin] = result; break; case FH_CENSURE: d_hessian[hidx_lin] = censure; case FH_RESULT_BIT_SET: if(Dxx+Dyy > 0.f) setLastBit(result); else clearLastBit(result); d_hessian[hidx_lin] = result; break; default: d_hessian[hidx_lin] = result; break; } } } void run_fasthessian_kernel(dim3 grid, dim3 threads, float * d_hessian, int octave) { fasthessian_kernel<<< grid, threads >>>(d_hessian, octave); } void run_eval_component_kernel(dim3 grid, dim3 threads, float * d_hessian, int octave, fh_component comp) { eval_component_kernel<<< grid, threads >>>(d_hessian, octave, (int)comp); } } // namespace asrl