Merge pull request #16 from patrick-lindemann/master

Fix deprecated np.int, np.float and np.object types

chesscog/corner_detection/detect_corners.py

@@ -103,7 +103,7 @@ def find_corners(cfg: CN, img: np.ndarray) -> np.ndarray:
     inverse_transformation_matrix = np.linalg.inv(transformation_matrix)
 
     # Warp grayscale image
-    dims = tuple(warped_img_size.astype(np.int))
+    dims = tuple(warped_img_size.astype(np.int32))
     warped = cv2.warpPerspective(gray, transformation_matrix, dims)
     borders = np.zeros_like(gray)
     borders[3:-3, 3:-3] = 1
@@ -122,7 +122,7 @@ def find_corners(cfg: CN, img: np.ndarray) -> np.ndarray:
     corners = np.array([[xmin, ymin],
                         [xmax, ymin],
                         [xmax, ymax],
-                        [xmin, ymax]]).astype(np.float)
+                        [xmin, ymax]]).astype(np.float32)
     corners = corners * scale
     img_corners = _warp_points(inverse_transformation_matrix, corners)
     img_corners = img_corners / img_scale
@@ -353,8 +353,8 @@ def _quantize_points(cfg: CN, warped_scaled_points: np.ndarray, intersection_poi
     mean_col_xs = warped_scaled_points[..., 0].mean(axis=0)
     mean_row_ys = warped_scaled_points[..., 1].mean(axis=1)
 
-    col_xs = np.rint(mean_col_xs).astype(np.int)
-    row_ys = np.rint(mean_row_ys).astype(np.int)
+    col_xs = np.rint(mean_col_xs).astype(np.int32)
+    row_ys = np.rint(mean_row_ys).astype(np.int32)
 
     # Remove duplicates
     col_xs, col_indices = np.unique(col_xs, return_index=True)
@@ -407,7 +407,7 @@ def _compute_vertical_borders(cfg: CN, warped: np.ndarray, mask: np.ndarray, sca
     G_x[~mask] = 0
 
     def get_nonmax_supressed(x):
-        x = (x * scale[0]).astype(np.int)
+        x = (x * scale[0]).astype(np.int32)
         thresh = cfg.BORDER_REFINEMENT.LINE_WIDTH // 2
         return G_x[:, x-thresh:x+thresh+1].max(axis=1)
 
@@ -431,7 +431,7 @@ def _compute_horizontal_borders(cfg: CN, warped: np.ndarray, mask: np.ndarray, s
     G_y[~mask] = 0
 
     def get_nonmax_supressed(y):
-        y = (y * scale[1]).astype(np.int)
+        y = (y * scale[1]).astype(np.int32)
         thresh = cfg.BORDER_REFINEMENT.LINE_WIDTH // 2
         return G_y[y-thresh:y+thresh+1].max(axis=0)
 

chesscog/occupancy_classifier/create_dataset.py

@@ -73,7 +73,7 @@ def warp_chessboard_image(img: np.ndarray, corners: np.ndarray) -> np.ndarray:
                            [BOARD_SIZE + SQUARE_SIZE, BOARD_SIZE + \
                             SQUARE_SIZE],  # bottom right
                            [SQUARE_SIZE, BOARD_SIZE + SQUARE_SIZE]  # bottom left
-                           ], dtype=np.float)
+                           ], dtype=np.float32)
     transformation_matrix, mask = cv2.findHomography(src_points, dst_points)
     return cv2.warpPerspective(img, transformation_matrix, (IMG_SIZE, IMG_SIZE))
 
@@ -84,7 +84,7 @@ def _extract_squares_from_sample(id: str, subset: str = "", input_dir: Path = RE
     with (input_dir / subset / (id + ".json")).open("r") as f:
         label = json.load(f)
 
-    corners = np.array(label["corners"], dtype=np.float)
+    corners = np.array(label["corners"], dtype=np.float32)
     unwarped = warp_chessboard_image(img, corners)
 
     board = chess.Board(label["fen"])

chesscog/piece_classifier/create_dataset.py

@@ -100,7 +100,7 @@ def warp_chessboard_image(img: np.ndarray, corners: np.ndarray) -> np.ndarray:
                            [BOARD_SIZE + MARGIN, \
                             BOARD_SIZE + MARGIN],  # bottom right
                            [MARGIN, BOARD_SIZE + MARGIN]  # bottom left
-                           ], dtype=np.float)
+                           ], dtype=np.float32)
     transformation_matrix, mask = cv2.findHomography(src_points, dst_points)
     return cv2.warpPerspective(img, transformation_matrix, (IMG_SIZE, IMG_SIZE))
 
@@ -111,7 +111,7 @@ def _extract_squares_from_sample(id: str, subset: str = "", input_dir: Path = RE
     with (input_dir / subset / (id + ".json")).open("r") as f:
         label = json.load(f)
 
-    corners = np.array(label["corners"], dtype=np.float)
+    corners = np.array(label["corners"], dtype=np.float32)
     unwarped = warp_chessboard_image(img, corners)
 
     board = chess.Board(label["fen"])

chesscog/recognition/recognition.py

@@ -108,7 +108,7 @@ def _classify_pieces(self, img: np.ndarray, turn: chess.Color, corners: np.ndarr
         pieces = self._pieces_model(piece_imgs)
         pieces = pieces.argmax(axis=-1).cpu().numpy()
         pieces = self._piece_classes[pieces]
-        all_pieces = np.full(len(self._squares), None, dtype=np.object)
+        all_pieces = np.full(len(self._squares), None, dtype=object)
         all_pieces[occupancy] = pieces
         return all_pieces
 

chesscog/report/prepare_confusion_matrix.py

@@ -54,7 +54,7 @@ def _get_category(piece: typing.Union[chess.Piece, None]) -> str:
 
 
 def _get_confusion_matrix(predicted: chess.Board, actual: chess.Board) -> np.ndarray:
-    matrix = np.zeros((len(CATEGORIES), len(CATEGORIES)), dtype=np.int)
+    matrix = np.zeros((len(CATEGORIES), len(CATEGORIES)), dtype=np.int32)
     for square in chess.SQUARES:
         pred = _get_category(predicted.piece_at(square))
         act = _get_category(actual.piece_at(square))
@@ -81,7 +81,7 @@ def _get_confusion_matrix(predicted: chess.Board, actual: chess.Board) -> np.nda
     # Filter out samples where the corners could not be detected
     df = df[(df["num_incorrect_corners"] != 4) | (df["error"] != "None")]
 
-    matrix = np.zeros((len(CATEGORIES), len(CATEGORIES)), dtype=np.int)
+    matrix = np.zeros((len(CATEGORIES), len(CATEGORIES)), dtype=np.int32)
     for i, row in df.iterrows():
         actual = chess.Board(row.fen_actual)
         predicted = chess.Board(row.fen_predicted)