Activity

1. Hemanath400 commented on Aug 30, 2026

   

   Hemanath400

   on Aug 30, 2026

   More actions

   I can reproduce this issue on Windows with Python 3.10.

   • OpenCV 4.13.0.92 → output shape (5, 1, 3)
   • OpenCV 5.0.0.93 → output shape (1, 5, 3)

   Using:

   import numpy as np
   import cv2
   
   x = np.array([0, 63, 127, 191, 255], dtype=np.uint8)
   colors = cv2.applyColorMap(x, cv2.COLORMAP_PARULA)
   print(colors.shape)

   The behavior change appears reproducible across versions.
2. ganlvtech commented on Sep 1, 2026

   

   ganlvtech

   on Sep 1, 2026

   Last edited by ganlvtech

   Author

   More actions

   Affected code: detectron2/DensePose

   https://github.com/facebookresearch/detectron2/blob/a2f4a8771ab77e8411c26b27f24f9489a28a2453/projects/DensePose/densepose/vis/densepose_results.py#L158

   class DensePoseResultsCustomContourVisualizer(DensePoseResultsVisualizer):
       """
       Contour visualization using marching squares
       """
   
       def __init__(self, levels=10, **kwargs):
           # TODO: colormap is hardcoded
           cmap = cv2.COLORMAP_PARULA
           if isinstance(levels, int):
               self.levels = np.linspace(0, 1, levels)
           else:
               self.levels = levels
           if "linewidths" in kwargs:
               self.linewidths = kwargs["linewidths"]
           else:
               self.linewidths = [1] * len(self.levels)
           self.plot_args = kwargs
           img_colors_bgr = cv2.applyColorMap((self.levels * 255).astype(np.uint8), cmap)
           self.level_colors_bgr = [
               [int(v) for v in img_color_bgr.ravel()] for img_color_bgr in img_colors_bgr
           ]
3. MDSALMANSHAMS commented on Sep 15, 2026

   

   MDSALMANSHAMS

   on Sep 15, 2026

   Contributor

   More actions

   Reproduced the shape difference too. Before this gets triaged as a bug — is the (1, N, 3) shape on 5.x an intentional layout change (e.g. matching a broader row/column-major convention update for 5.x) or a genuine regression? It's already breaking downstream consumers that assume the 4.x (N, 1, 3) shape (detectron2/DensePose per the comment above). If unintentional, could this get an actionable label? The fix would live in opencv/opencv core (imgproc), not this packaging repo, so happy to help there once it's confirmed which side of that line it's on.

   👍React with 👍1Reacted by Ganlv