Neural net magicness

main.py

@@ -24,9 +24,15 @@ def identify_piece(image: np.ndarray, position: POSITION, sift: cv2.xfeatures2d_
     for piece in OUR_PIECES:
         for color in COLOR:
             #color = runner.compute_color(file, rank)
+            #classifier = load_classifier(f"classifiers/neural_net_{piece}/{color}.pkl")
             classifier = load_classifier(f"classifiers/classifier_{piece}/{color}.pkl")
             features = runner.generate_bag_of_words(image, centers, sift)
             prob = classifier.predict_proba(features)
+
+            #image = cv2.resize(image, (172, 172))
+            #data = np.reshape(image, (1, np.product(image.shape)))
+            #prob = classifier.predict_proba(data)
+
             probs[piece.name][color.name] = prob[0, 1]
             print(f"{piece}, {color}, {prob[0, 1]}")
             #if prob[0, 1] > best and color == position.color:  # can only be best if correct color. Iterating through both colors for debugging only
@@ -102,9 +108,9 @@ def remove_most_empties(warped):
     for position in POSITION:
         counter = 0
 
-        src = runner.get_square(warped, position)
+        img_src = runner.get_square(warped, position)
-        width, height, _ = src.shape
+        width, height, _ = img_src.shape
-        src = src[width // 25:, height // 25:]
+        src = img_src[width // 25:, height // 25:]
         # src = src[:-width//200, :-height//200]
         segmentator = cv2.ximgproc.segmentation.createGraphSegmentation(sigma=0.8, k=150, min_size=700)
         segment = segmentator.processImage(src)
@@ -116,16 +122,23 @@ def remove_most_empties(warped):
             masked.mask = mask != i
 
             y, x = np.where(segment == i)
+
+
+            if position == POSITION.E8:
+                print(np.max(segment))
+                print(len(y))
+
+
             pls.append(len(y))
             top, bottom, left, right = min(y), max(y), min(x), max(x)
 
             dst = masked.filled()[top: bottom + 1, left: right + 1]
-            cv2.imwrite(f"segment_test/segment_{datetime.utcnow().timestamp()}_{position}.png", dst)
+            cv2.imwrite(f"tmp_seg/segment_{datetime.utcnow().timestamp()}_{position}.png", dst)
 
         if np.max(segment) > 0 and not np.all([x < (164 ** 2) * 0.2 for x in pls]) and (
-                np.max(segment) >= 3 or np.all([x < (164 ** 2) * 0.942 for x in pls])):
+                np.max(segment) >= 3 or np.all([x < (164 ** 2) * 0.9469 for x in pls])):
             #print(f"{position} is nonempty")
-            non_empties.append([position, src])
+            non_empties.append([position, img_src])
             empty += 1
 
     #print(64 - empty)
@@ -148,21 +161,29 @@ if __name__ == '__main__':
 
     #runner.train_pieces_svm()
 
-    #board = cv2.imread("whole_boards/boards_for_empty/board_1554286488.605142_rank_3.png")
+    board = cv2.imread("whole_boards/boards_for_empty/board_1554288891.129901_rank_8.png")
-    board = cv2.imread("whole_boards/boards_for_empty/board_1554286515.323962_rank_3.png")
+    #board = cv2.imread("whole_boards/boards_for_empty/board_1554286515.323962_rank_3.png")
     warped = runner.warp_board(board)
 
+    tmp = find_occupied_squares(warped)
+    for pos, square in tmp:
+        cv2.imshow(f"{pos}", square)
+    cv2.waitKey(0)
+    exit()
+
+    """
     rook_square = runner.get_square(warped, POSITION.H3)
     knight_square = runner.get_square(warped, POSITION.D3)
     cv2.imshow("lel", rook_square)
     cv2.imshow("lil", knight_square)
-    rook_out = cv2.Canny(rook_square, 50, 55, L2gradient=True)
+    #rook_out = cv2.Canny(rook_square, 50, 55, L2gradient=True)
     knight_out = cv2.Canny(knight_square, 50, 55, L2gradient=True)
-    cv2.imshow("lal", rook_out)
+    knight_out_l = cv2.Canny(knight_square, 50, 55, L2gradient=False)
-    cv2.imshow("lul", knight_out)
+    cv2.imshow("lal", knight_out)
+    cv2.imshow("lul", knight_out_l)
     cv2.waitKey(0)
     exit()
-
+    """
     occupied = find_occupied_squares(warped)
 
     sift = cv2.xfeatures2d.SIFT_create()

runner.py

@@ -9,6 +9,7 @@ from sklearn import cluster, metrics, svm
 from sklearn.externals import joblib
 from sklearn.pipeline import make_pipeline
 from sklearn.preprocessing import StandardScaler
+from sklearn import neural_network
 
 from util import RANK, POSITION, imwrite, PIECE, COLOR, Squares, OUR_PIECES
 
@@ -108,11 +109,28 @@ def train_pieces_svm() -> None:
             current_weight = len(glob.glob(f"training_images/{piece}/{color}_square/*.png"))
             print(f"Training for piece: {piece}")
             X, Y = load_training_data(piece, color)
-            classifier = svm.SVC(C=10, gamma=0.01, class_weight={0: 45, 1: 1}, probability=True)
+            classifier = svm.SVC(C=10, gamma=0.01, class_weight={0: 15, 1: 0.8}, probability=True)
             classifier.fit(X, Y)
             joblib.dump(classifier, f"classifiers/classifier_{piece}/{color}.pkl")
 
 
+def train_pieces_svm_canny() -> None:
+    for square_color in COLOR:
+        X = []
+        Y = []
+        for piece in OUR_PIECES + (PIECE.EMPTY,):
+            for filename in glob.glob(f"training_images/{piece}/{square_color}_square/*.png"):
+                img = cv2.imread(filename)
+                y, x = np.histogram(img.ravel(), bins=32, range=[0, 256])
+                X.append(y)
+                Y.append(piece == PIECE.EMPTY)
+
+        classifier = make_pipeline(StandardScaler(),
+                                   svm.SVC(C=10.0, gamma=0.01, probability=True))
+        classifier.fit(np.array(X), np.array(Y))
+        joblib.dump(classifier, f"classifiers/classifier_empty/white_piece_on_{square_color}_square.pkl")
+
+
 def warp_board(camera_image, debug_image=None) -> np.ndarray:
     baseline = cv2.imread("new_baseline_board.png")
 
@@ -181,10 +199,10 @@ def get_square(warped_board: np.ndarray, position: POSITION) -> np.ndarray:
     side = int(width * 0.045)
     size = width - 2 * side
     square_size = size // 8
-    padding = 0
+    padding = 2
 
     x1 = side + (square_size * (position.file - 1))
-    x2 = x1 + square_size
+    x2 = x1 + square_size + padding
     y1 = max(0, side + (square_size * (8 - position.rank)) - padding)  # 8 - rank because chessboard is from 8 to 1
 
     y2 = min(width, y1 + square_size + padding)
@@ -207,5 +225,47 @@ def save_empty_fields(warped_board: np.ndarray, skip_rank: RANK = None) -> None:
         imwrite(f"training_images/empty/{position.color}_square/training_{position}_{datetime.utcnow().timestamp()}.png", square)
 
 
+def load_data_nn(spec_piece, color):
+    X = None
+    Y = None
+    for piece in OUR_PIECES:
+        piece_class = int(spec_piece == piece)
+
+        for filename in glob.glob(f"training_images/{piece}/{color}_square/*.png"):
+            image = cv2.imread(filename)
+            data = np.reshape(image, (1, np.product(image.shape)))
+            if X is None:
+                if piece_class == 1:
+                    for _ in range(10):
+                        X = np.array(data)
+                        Y = np.array([piece_class])
+                else:
+                    for _ in range(5):
+                        X = np.array(data)
+                        Y = np.array([piece_class])
+            else:
+                if piece_class == 1:
+                    for _ in range(10):
+                        X = np.vstack((X, data))
+                        Y = np.vstack((Y, [piece_class]))
+                else:
+                    for _ in range(5):
+                        X = np.vstack((X, data))
+                        Y = np.vstack((Y, [piece_class]))
+    return X, Y
+
+
+def train_nn():
+    for piece in OUR_PIECES:
+        for color in COLOR:
+            X, Y = load_data_nn(piece, color)
+
+            classifier = neural_network.MLPClassifier(hidden_layer_sizes=256)
+
+            classifier.fit(X, Y)
+
+            joblib.dump(classifier, f"classifiers/neural_net_{piece}/{color}.pkl")
+
+
 if __name__ == '__main__':
-    train_empty_or_piece_hist()
+    train_nn()

tensor_classifier.py

@@ -0,0 +1,72 @@
+from __future__ import absolute_import, division, print_function, unicode_literals
+
+import tensorflow as tf
+import tensorflow_datasets as tfds
+from tensorflow.python.keras import datasets, layers, models
+import glob
+import numpy as np
+import cv2
+
+import runner
+from main import find_occupied_squares
+from util import POSITION
+
+new_model = models.load_model('test_chess_model.h5')
+new_model.summary()
+
+#board = cv2.imread("whole_boards/boards_for_empty/board_1554286488.605142_rank_3.png")
+#board = cv2.imread("whole_boards/boards_for_empty/board_1554285167.655788_rank_5.png")
+board = cv2.imread("whole_boards/boards_for_empty/board_1554288891.129901_rank_8.png")
+
+
+warped = runner.warp_board(board)
+
+occupied = find_occupied_squares(warped)
+
+pos_1 = POSITION.C5
+pos_2 = POSITION.D5
+pos_3 = POSITION.G5
+pos_4 = POSITION.H5
+
+square_1 = runner.get_square(warped, pos_1)
+square_2 = runner.get_square(warped, pos_2)
+square_3 = runner.get_square(warped, pos_3)
+square_4 = runner.get_square(warped, pos_4)
+
+square_1 = cv2.cvtColor(square_1, cv2.COLOR_BGR2GRAY)
+square_2 = cv2.cvtColor(square_2, cv2.COLOR_BGR2GRAY)
+square_3 = cv2.cvtColor(square_3, cv2.COLOR_BGR2GRAY)
+square_4 = cv2.cvtColor(square_4, cv2.COLOR_BGR2GRAY)
+
+width, height = square_1.shape
+square_1 = square_1 / 255.0
+square_2 = square_2 / 255.0
+square_3 = square_3 / 255.0
+square_4 = square_4 / 255.0
+
+
+pieces = ['knight', 'rook']
+
+for pos, square in occupied:
+    square = cv2.cvtColor(square, cv2.COLOR_BGR2GRAY)
+    width, height = square.shape
+    square = square / 255.0
+    test = new_model.predict(np.array(square).reshape((-1, width, height, 1)))
+    text_color = 255
+    cv2.putText(square, f"{pos} {pieces[int(np.argmax(test))]}", (0, 50), cv2.FONT_HERSHEY_SIMPLEX, fontScale=1,
+                color=(text_color,) * 3, thickness=3)
+    cv2.imshow(f"{pos}", square)
+
+cv2.waitKey(0)
+
+"""
+for pos, square in [(pos_1, square_1), (pos_2, square_2), (pos_3, square_3), (pos_4, square_4)]:
+    test = new_model.predict(np.array(square).reshape((-1, width, height, 1)))
+    print(f"{pos}: {np.argmax(test)}")
+    text_color = 255
+    cv2.putText(square, f"{pos} {pieces[int(np.argmax(test))]}", (0, 50), cv2.FONT_HERSHEY_SIMPLEX, fontScale=1,
+                color=(text_color,) * 3, thickness=3)
+    cv2.imshow(f"{pos}", square)
+    """
+
+