Fixes ?

Panjaksli · Aug 31, 2023 · 00a82de · 00a82de
1 parent f92dc07
commit 00a82de
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 6 deletions.
diff --git a/BNN/NNet/NNet.cpp b/BNN/NNet/NNet.cpp
@@ -118,6 +118,7 @@ namespace BNN {
 				println("Failed training the network !!!");
 				return false;
 			}
+			cost *= mult;
 			net.Zero();
 			for(const auto& n : nets) {
 				for(idx i = 1; i < net.graph.size() - 1; i++) {
@@ -131,10 +132,10 @@ namespace BNN {
 					if(net.optimizer->get_mb(i)) *net.optimizer->get_mb(i) += mult * (*n.optimizer->get_mb(i));
 				}
 			}
-			printr("Step:", step + 1, "Cost:", cost * mult, "Time:", timer(t), "                        ");
+			best_cost = fminf(cost, best_cost);
+			printr("Step:", step + 1, "Cost:", cost, "Time:", timer(t), "                    ");
 		}
-		best_cost = fminf(cost * mult, best_cost);
-		println("Trained Network:", name, "Cost:", cost * mult, "Best:", best_cost, "Time:", timer(t), "                        ");
+		println("Trained Network:", name, "Cost:", cost, "Best:", best_cost, "Time:", timer(t), "                ");
 		*this = net;
 		return true;
 	}

diff --git a/Example1/main.cpp b/Example1/main.cpp
@@ -3,7 +3,7 @@ using namespace BNN;
 
 //rgb upscaling
 int main() {
-#if 0
+#if 1
 	//Inference code
 	NNet net("Upscaling_net");
 	std::string inp, outp;
@@ -21,7 +21,7 @@ int main() {
 			y.chip(i, 0) = net.Compute_DS(x.chip(i, 0).reshape(dim1<3>{1, x.dimension(1), x.dimension(2)})).chip(i, 0);
 		Image(y).save(outp + ".png");*/
 		//DIFF MODEL
-		Tensor x = Image(inp, 3, 1).tensor_rgb();
+		Tensor x = Image(inp, 3, 1).tensor_rgb(true);
 		Image(net.Compute_DS(x) + resize(x, 2, 2)).save(outp + ".png");
 		//Cubic
 		//Image(inp, 3, 1).resize_cubic(960,720).save(outp + ".png");
@@ -33,7 +33,7 @@ int main() {
 	std::string parent = "Upscaler/";
 	//Input data
 	std::string in_folder = "Downscaler/";
-	std::string netname = parent + "ups_c5x32_c3x32x3_c3_ps2_dy_cu";
+	std::string netname = parent + "ups_c5x32_c3x32x2_c3_ps2_dy_cu";
 	Tenarr x(3, 240, 160, train_set);
 #pragma omp parallel for
 	for(idx i = 0; i < train_set; i++)