Merge branch 'master' of git.stim.ee.uh.edu:codebase/stimlib

David Mayerich
2 parents ee52e24f d9a27f16
Showing 2 changed files with 26 additions and 48 deletions Show diff stats
stim/iVote/ivote2.cuh
stim/iVote/ivote2/local_max.cuh
@@ -94,8 +94,7 @@ namespace stim {
 	//this function performs the 2D iterative voting algorithm on the image stored in the gpu 
 	template<typename T>
-	void gpu_ivote2(T* gpuI, unsigned int rmax, size_t x, size_t y, bool invert = false, T t = 0, std::string outname_img = "out.bmp", std::string outname_txt = "out.txt",
-					int iter = 8, T phi = 15.0f * (float)stim::PI / 180, int conn = 8, bool debug = false) {
+	void gpu_ivote2(T* gpuI, unsigned int rmax, size_t x, size_t y, bool invert = false, T t = 0, int iter = 8, T phi = 15.0f * (float)stim::PI / 180, int conn = 8, bool debug = false) {
 		size_t pixels = x * y;				//compute the size of input image
 		//
@@ -129,43 +128,36 @@ namespace stim {
 		}
 		stim::cuda::gpu_local_max<float>(gpuI, gpuVote, conn, x, y);				//calculate the local maxima
-		T* pts = (T*)malloc(bytes);													//allocate memory on the cpu to store the output of iterative voting
-		HANDLE_ERROR(cudaMemcpy(pts, gpuI, bytes, cudaMemcpyDeviceToHost));			//copy the output from gpu to the cpu memory
-		
-		T threshold;
-		if (t == 0) threshold = stim::th_otsu<T>(pts, pixels);	//if threshold value is not set call the function to compute the threshold
-		else threshold = t;
-		
-		std::ofstream output;		//save the thresholded detected seeds in a text file
-		output.open(outname_txt);
-		output << "X" << " " << "Y" << " " << "threshold" << "\n";
-		size_t ind;
-		for (size_t ix = 0; ix < x; ix++) {
-			for (size_t iy = 0; iy < y; iy++) {
-				ind = iy * x + ix;
-				if (pts[ind] > threshold) {
-					output << ix << " " << iy << " " << pts[ind] << "\n";
-					pts[ind] = 1;
+		if (t > 0) {
+			T* pts = (T*)malloc(bytes);													//allocate memory on the cpu to store the output of iterative voting
+			HANDLE_ERROR(cudaMemcpy(pts, gpuI, bytes, cudaMemcpyDeviceToHost));			//copy the output from gpu to the cpu memory
+
+			T threshold;
+			threshold = t;
+
+			size_t ind;
+			for (size_t ix = 0; ix < x; ix++) {
+				for (size_t iy = 0; iy < y; iy++) {
+					ind = iy * x + ix;
+					if (pts[ind] > threshold) {
+						pts[ind] = 1;
+					}
+					else pts[ind] = 0;
 				}
-				else pts[ind] = 0;
 			}
+			HANDLE_ERROR(cudaMemcpy(gpuI, pts, bytes, cudaMemcpyHostToDevice));		//copy the points to the gpu
 		}
-		output.close();
-
-		HANDLE_ERROR(cudaMemcpy(gpuI, pts, bytes, cudaMemcpyHostToDevice));		//copy the points to the gpu
-		stim::cpu2image(pts, outname_img, x, y); //output the image
-		
+				
 	}
 	template<typename T>
-	void cpu_ivote2(T* cpuI, unsigned int rmax, size_t x, size_t y, bool invert = false, T t = 0, std::string outname_img = "out.bmp", std::string outname_txt = "out.txt",
-					int iter = 8, T phi = 15.0f * (float)stim::PI / 180, int conn = 8, bool debug = false) {
+	void cpu_ivote2(T* cpuI, unsigned int rmax, size_t x, size_t y, bool invert = false, T t = 0, int iter = 8, T phi = 15.0f * (float)stim::PI / 180, int conn = 8, bool debug = false) {
 		size_t bytes = x*y * sizeof(T);
 		T* gpuI;						//allocate space on the gpu to save the input image
 		HANDLE_ERROR(cudaMalloc(&gpuI, bytes));
 		HANDLE_ERROR(cudaMemcpy(gpuI, cpuI, bytes, cudaMemcpyHostToDevice));		//copy the image to the gpu
-		stim::gpu_ivote2<T>(gpuI, rmax, x, y, invert, t, outname_img, outname_txt, iter, phi, conn, debug);				//call the gpu version of the ivote
+		stim::gpu_ivote2<T>(gpuI, rmax, x, y, invert, t, iter, phi, conn, debug);				//call the gpu version of the ivote
 		HANDLE_ERROR(cudaMemcpy(cpuI, gpuI, bytes, cudaMemcpyDeviceToHost));		//copy the output to the cpu
 	}
 }
@@ -10,30 +10,22 @@ namespace stim{
 		// this kernel calculates the local maximum for finding the cell centers
 		template<typename T>
-		__global__ void cuda_local_max(T* gpuCenters, T* gpuVote, int conn, size_t x, size_t y){
+		__global__ void cuda_local_max(T* gpuCenters, T* gpuVote, int conn, int x, int y){
 			// calculate the 2D coordinates for this current thread.
-			size_t xi = blockIdx.x * blockDim.x + threadIdx.x;
-			size_t yi = blockIdx.y * blockDim.y + threadIdx.y;
+			int xi = blockIdx.x * blockDim.x + threadIdx.x;
+			int yi = blockIdx.y * blockDim.y + threadIdx.y;
 			if(xi >= x || yi >= y)
 				return;
 			// convert 2D coordinates to 1D
-			size_t i = yi * x + xi;
+			int i = yi * x + xi;
 			gpuCenters[i] = 0;		//initialize the value at this location to zero
 			T val = gpuVote[i];
-			//compare to the threshold
-			//if(val < final_t) return;
-			
-			//define an array to store indices with same vote value
-			/*int * IdxEq;
-			IdxEq = new int  [2*conn];
-			int n = 0;*/
-			
 			for(int xl = xi - conn; xl < xi + conn; xl++){
 				for(int yl = yi - conn; yl < yi + conn; yl++){
 					if(xl >= 0 && xl < x && yl >= 0 && yl < y){
@@ -42,8 +34,7 @@ namespace stim{
 							return;
 							}
 						if (gpuVote[il] == val){
-							/*IdxEq[n] = il;
-							n = n+1;*/
+							
 							 if( il > i){
 								 return;
 							}
@@ -51,12 +42,7 @@ namespace stim{
 					}							
 				}
 			}
-			/*if (n!=0){
-				if(IdxEq[n/2] !=i){
-					return;
-				}
-			}	*/	
-			//gpuCenters[i] = 1;
+			
 			gpuCenters[i] = gpuVote[i];
 		}