I am trying to detect the number of certain objects in an image using Custom Vision. The images I am sending for predictions are thin and long, with aspect ratios of typically 1:12. However, once the predictions are made, the top 10-15% of the image height is left unpredicted upon while there are clear and unobstructed objects remaining in the image.
I am seeing this consistently across the board with images of aspect ratio 1:8 or higher. Pixel heights have not exceeded the 10,240p limit, nor is the image size above the 4MB cap.
At the moment, I am needing to cut the image in half to get predictions for the whole image but would like to move away from that (as it causes me to occasionally divide objects and doubles my api calls). Does anyone know why I might be running into this error?
