Q1, Q2: Featured extraction done and prior probability estimation done.

README.md

@@ -2,7 +2,7 @@
 **Dealine**: 01.04.2021
 
 Please put your name here:  
-**Name:** .......
+**Name:** Chirag Vaghela, Akarawint Chawalitanont
 ## Foreword
 ### Implementation of a Minimal Classification System
 

src/Bayes.cpp

@@ -14,13 +14,13 @@ CBayes::CBayes(byte nStates, word nFeatures)
 void CBayes::addFeatureVec(const Mat & featureVector, byte gt)
 {
 	// --- PUT YOUR CODE HERE ---
-
-
-
+
 	for (word f = 0; f < m_nFeatures; f++) {
 		byte feature = featureVector.at<byte>(f, 0);
 		m_vPDF[f * m_nStates + gt]->addPoint(feature);
 	}
+
+    m_pPrior->addPoint(gt);
 }
 
 Mat CBayes::getPotentials(const Mat & featureVector) const
@@ -30,15 +30,15 @@ Mat CBayes::getPotentials(const Mat & featureVector) const
 	for (byte s = 0; s < m_nStates; s++)
 		res.at<float>(s, 0) = static_cast<float>(m_pPrior->getDensity(s));
 
-	// --- PUT YOUR CODE HERE ---
+    for (byte s = 0; s < m_nStates; s++)
+        for (word f = 0; f < m_nFeatures; f++)
+            res.at<float>(s, 0) = m_vPDF[f * m_nStates + s]->isEstimated() ? res.at<float>(s, 0) * m_vPDF[f * m_nStates + s]->getDensity(featureVector.at<byte>(f, 0)): 0;
 
-
-
-	return res;
+    return res;
 }
 
 void CBayes::printPriorProbabilities(void) {
 	for (byte s = 0; s < m_nStates; s++)
 		printf("%.1f%%\t", 100 * m_pPrior->getDensity(s));
 	printf("\n");
-}
+}

src/main.cpp

@@ -33,26 +33,33 @@ int main(int argc, char *argv[])
 	for (int y = 0; y < imgSize.height; y++) {
 		for (int x = 0; x < imgSize.width; x++) {
 			// --- PUT YOUR CODE HERE ---
-
+            featureVector.row(0) = train_fv.at<Vec3b>(y, x)[0];
+            featureVector.row(1) = train_fv.at<Vec3b>(y, x)[1];
+            featureVector.row(2) = train_fv.at<Vec3b>(y, x)[2];
 
 
 			if (x == 0 && y == 0)
 				std::cout << featureVector << std::endl;
 
 			byte gt = train_gt.at<byte>(y, x);
+
 			classifier->addFeatureVec(featureVector, gt);
 		} // x
 	} // y
 	Timer::stop();
 
 	classifier->printPriorProbabilities();
+    //    17.2%    0.4%    59.5%    9.9%    13.0%    0.0%
+    //    Car class is not represented in training image because 0.0% probability
 
 	// ========================= Testing =========================
 	Timer::start("Testing... ");
 	for (int y = 0; y < imgSize.height; y++) {
 		for (int x = 0; x < imgSize.width; x++) {
 			// --- PUT YOUR CODE HERE ---
-
+            featureVector.row(0) = test_fv.at<Vec3b>(y, x)[0];
+            featureVector.row(1) = test_fv.at<Vec3b>(y, x)[1];
+            featureVector.row(2) = test_fv.at<Vec3b>(y, x)[2];
 
 
 			// get potentials
@@ -61,21 +68,39 @@ int main(int argc, char *argv[])
 			// find the largest potential
 			byte classLabel = 0;
 			// --- PUT YOUR CODE HERE ---
-
-
-
+            double minVal;
+            double maxVal;
+            Point minLoc;
+            Point maxLoc;
+            minMaxLoc( potentials, &minVal, &maxVal, &minLoc, &maxLoc );
+//            float maxV = 0.0;
+//            float maxL = 0;
+//            for (int i = 0; i < potentials.rows; i++) {
+//                std::cout << potentials.at<float>(i) << " ";
+//                if (potentials.at<float>(i) > maxV) {
+//                    maxV = potentials.at<float>(i);
+//                    maxL = i;
+//                }
+//            }
+//            std::cout << "MaxV:" << maxVal <<  "maxL:" << " " << maxLoc.y;
+
+            classLabel = (byte)maxLoc.y;
 			solution.at<byte>(y, x) = classLabel;
 		}
+//        cout << "\n";
 	}
 	Timer::stop();
 
 	// ====================== Evaluation =======================	
 	Timer::start("Evaluation... ");
 	double accuracy = 0;
-	for (int y = 0; y < imgSize.height; y++)
-		for (int x = 0; x < imgSize.width; x++)
-			if (solution.at<byte>(y, x) == test_gt.at<byte>(y, x))
-				accuracy++;
+    for (int y = 0; y < imgSize.height; y++) {
+        for (int x = 0; x < imgSize.width; x++){
+            if (solution.at<byte>(y, x) == test_gt.at<byte>(y, x)) {
+                accuracy++;
+            }
+        }
+    }
 	accuracy /= (imgSize.height * imgSize.width);
 	Timer::stop();
 	printf("Accuracy = %.2f%%\n", accuracy * 100);
@@ -87,6 +112,10 @@ int main(int argc, char *argv[])
 	imshow("Test image", test_img);
 	imshow("groundtruth", test_gt);
 	imshow("solution", solution);
+
+    imwrite(dataPath + "renders/test_img.png", test_img);
+    imwrite(dataPath + "renders/test_gt.png", test_gt);
+    imwrite(dataPath + "renders/solution.png", solution);
 	waitKey();
 
 	return 0;