Compute winding order of triangles

quantized_mesh_encoder/util.py

@@ -13,3 +13,42 @@ def zig_zag_encode(arr):
     int16
     """
     return np.bitwise_xor(np.right_shift(arr, 15), np.left_shift(arr, 1))
+
+
+def is_ccw(positions, indices):
+    """For each triangle, determine if in counter-clockwise winding order
+
+    Implements algorithm from here:
+    https://stackoverflow.com/a/1165943
+
+    NOTE: this algorithm is reversed if the origin is the top left (i.e. as in a
+    PNG), rather than the standard origin of the bottom left.
+
+    Args:
+        positions: numpy array of shape (-1, 3) defining positions. Each row is a single coordinate
+        indices: numpy array either flat or of shape (-1, 3) defining triangles
+
+    Returns:
+        1d numpy array where each value represents a single triangle formed by a
+        triple within `indices`. The value is `True` if the triangle is
+        counter-clockwise or `False` if clockwise.
+    """
+    # Make sure indices
+    indices = indices.reshape(-1, 3)
+
+    # Perform coordinate lookup in positions using indices
+    # `coords` is then an array of shape (-1, 3, 3) where each block of (i, 3,
+    # 3) represents all the coordinates of a single triangle
+    coords = positions[indices]
+
+    # x2 - x1 for every edge of every triangle
+    x2x1 = np.roll(coords[:, :, 0], -1, axis=1) - coords[:, :, 0]
+
+    # y2 + y1 for every edge of every triangle
+    y2y1 = np.roll(coords[:, :, 1], -1, axis=1) + coords[:, :, 1]
+
+    # Sum edge weights for each triangle
+    axissum = (x2x1 * y2y1).sum(axis=1)
+
+    # Values <0 represent counter-clockwise triangles
+    return axissum < 0