Manao
12th February 2004, 21:53
I am currently making a motion vector toolbox for avisynth, which will compute motion vectors on a clip.
I already have something that is working, and I would like to know some figures, in order to compare the efficiency of my search against the search of XviD or any other motion estimation algorithm that you may have tested.
In order to evaluate the efficiency, I compute the PSNR between the frame and the frame motion compensated. When I do so, I obtain a PSNR around 40 dB ( with 8x8 blocks, half pel search, a clip resolution of 712 x 328 ) for low motion ( 2 - 3 pixels at most ), 35 dB for medium motion ( 3 - 8 ) and 30 dB and lower for high motion.
I know it will depend on the quality of my source ( DVD, but bad quality : it's titanic, but in a 4/3 DVD ( ouch ), and a double layer mono face DVD with 3 languages ), but is it sufficient for you to tell me whether the figures are good or not ?
Else, where can I find reference clips in order to test my algorithm ?
I already have something that is working, and I would like to know some figures, in order to compare the efficiency of my search against the search of XviD or any other motion estimation algorithm that you may have tested.
In order to evaluate the efficiency, I compute the PSNR between the frame and the frame motion compensated. When I do so, I obtain a PSNR around 40 dB ( with 8x8 blocks, half pel search, a clip resolution of 712 x 328 ) for low motion ( 2 - 3 pixels at most ), 35 dB for medium motion ( 3 - 8 ) and 30 dB and lower for high motion.
I know it will depend on the quality of my source ( DVD, but bad quality : it's titanic, but in a 4/3 DVD ( ouch ), and a double layer mono face DVD with 3 languages ), but is it sufficient for you to tell me whether the figures are good or not ?
Else, where can I find reference clips in order to test my algorithm ?