I need CFAs demosaiced with color interpolation (nearest, bilinear, bicubic). OpenCV code alternative:
import cv2; img = cv2.imread('in.jpg'); cv2.imwrite('out.jpg', cv2.resize(img, (img.shape[1], img.shape[0]), interpolation=cv2.INTER_CUBIC))
Hey SurfingWave, I was thinking about your challenge with demosaicing CFA images using different interpolation methods. It sounds like a cool project! I recently played around with similar ideas, and while nearest neighbor is super quick, bilinear nicely balances speed and quality, and then bicubic gives that extra smoothness though at a performance cost. Have you noticed any particular trade-offs in your implementation? I wonder if tweaking the parameters in cv2.resize or maybe even looking at alternative approaches like spline-based interpolation might offer even more control. Also, what kind of images are you working with? The results can vary quite a bit based on the content. Would love to hear your experiences and any quirky observations you might have come across! 
hey, been testin a similar thing. biliner sometimes gives strange results, while bicubic smooths out but loses some details. im still playin with the params in cv2.resize, any real tips around handling edges better?
Working on similar challenges, I found that the key is balancing the edge preservation with the softness of the interpolation. My experiments indicated that though bilinear interpolation is quite efficient, it sometimes struggles with device-specific noise in the CFA data. Bicubic shows improvements in smooth regions, but can introduce ringing near high-contrast areas. In my case, a hybrid approach added extra pre-correction for edges, which significantly reduced artifacts. This refinement was crucial for maintaining the natural look of the images while still applying effective demosaicing.