When training and testing my network I noticed that the resolution of the recorded images (1080x160) is very different from that we get during testing(1440x320). But not only the resolution but also the aspect ratio is different, which leads to very bad predictions even when rescaling the image. Could we somehow match those resolutions or should we just manually fix them by cropping & rescaling?
When training and testing my network I noticed that the resolution of the recorded images (1080x160) is very different from that we get during testing(1440x320). But not only the resolution but also the aspect ratio is different, which leads to very bad predictions even when rescaling the image. Could we somehow match those resolutions or should we just manually fix them by cropping & rescaling?