I have a 1440x1080 video with SAR 4:3 and DAR 16:9 recorded with a Sony α6000. Having DAR 16:9 when viewing the video it appears correctly, a horizontal stretching is automatically performed and shown as if it were 1920x1080. However, when extracting the frames (with opencv) with the native resolution they appear squashed (4:3) probably because the displayed image is not interpreted correctly. If these video frames are to be used for training or for inference, should a horizontal stretch be performed and then a resize of the images to 1920x1080 or alternatively a resize of the video directly, or should they be used with their native resolution 1440x1080? Continue reading...
