added slicing of image to confirm to stride length

Signed-off-by: Yuri van Warmerdam <[email protected]>

rcdt_detection/nodes/object_detection.py

@@ -19,7 +19,7 @@
 from sensor_msgs.msg import Image, CameraInfo
 
 from rcdt_detection.image_manipulation import (
-    ros_image_to_cv2_image,
+    ros_image_to_cv2_image_sliced,
     segmentation_mask_to_binary_mask,
     single_to_three_channel,
 )
@@ -146,10 +146,8 @@ def process_rgb_image(
     If it is, the image coordinates of its centroid are calculated.
 
     """
-    rgb_image = ros_image_to_cv2_image(message)
-
+    rgb_image = ros_image_to_cv2_image_sliced(message)
     segmentation_result = segment_image(model=segmentation_model, image=rgb_image)
-    ros_logger.info(f"Segmented {len(segmentation_result.masks)} objects.")
 
     centroid_image_coordinates = []
     for mask in segmentation_result.masks:
@@ -171,7 +169,7 @@ def process_depth_image(
     intrinsics: rs2.intrinsics,
 ) -> list[Point]:
     """Calculate world coordinate relative to the camera of image coordinates."""
-    depth_image = ros_image_to_cv2_image(message)
+    depth_image = ros_image_to_cv2_image_sliced(message)
 
     world_coordinates = []
     for image_coordinate in image_coordinates:

rcdt_detection/rcdt_detection/image_manipulation.py

@@ -3,29 +3,51 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import cv2
-import numpy as np
+from numpy import array, uint8
 import torch
 from sensor_msgs.msg import Image
 from cv_bridge import CvBridge
+from math import floor, ceil
 
 cv_bridge = CvBridge()
 
 
 def ros_image_to_cv2_image(
     image_message: Image, desired_encoding: str = "passthrough"
-) -> np.array:
+) -> array:
     """Convert ROS image message to cv2 image."""
     return cv_bridge.imgmsg_to_cv2(image_message, desired_encoding=desired_encoding)
 
 
-def segmentation_mask_to_binary_mask(mask: torch.Tensor) -> np.array:
+def slice_image_to_stride(image: array, stride: int = 32) -> array:
+    """Slice image to confirm to a given stride length"""
+    rows = image.shape[0]
+    cols = image.shape[1]
+    row_dist = rows % stride
+    col_dist = cols % stride
+    return image[
+        slice(floor(row_dist / 2), rows - ceil(row_dist / 2)),
+        slice(floor(col_dist / 2), cols - ceil(col_dist / 2)),
+    ]
+
+
+def ros_image_to_cv2_image_sliced(
+    image_message: Image, desired_encoding: str = "passthrough", stride: int = 32
+) -> array:
+    """slice an image so that its dimensions are a multiple of stride"""
+    return slice_image_to_stride(
+        ros_image_to_cv2_image(image_message, desired_encoding), stride
+    )
+
+
+def segmentation_mask_to_binary_mask(mask: torch.Tensor) -> array:
     """Convert given mask to np.array with range [0, 255], dtype=uint8, and dimensions [height, width, channels]."""
-    binary_mask = mask.data.cpu().numpy().astype(np.uint8)
+    binary_mask = mask.data.cpu().numpy().astype(uint8)
     binary_mask = binary_mask * 255
     binary_mask = binary_mask.transpose(1, 2, 0)
     return binary_mask
 
 
-def single_to_three_channel(image: np.array) -> np.array:
+def single_to_three_channel(image: array) -> array:
     """Convert given single-channel image to three-channel image."""
     return cv2.cvtColor(image, cv2.COLOR_GRAY2BGR)