Where can I get a program that outputs an "I-frame only" MPEG-2 video?

[Videogamer555]

I know that occording to Wiki there's MPEG-2 video streams that contain I, P, and B frames and MPEG-2 video streams that contain only I and P frames. However I also recall reading somewhere that the MPEG-2 standard also supports MPEG-2 video streams that consist ONLY of I-frames. So I'm wondering what video conversion software exists that supports outputting I-frame only MPEG-2 videos?

[El Heggunte]

If you set the encoder to use a GOP-length = 1 and zero B-frames, then you'll get a video stream containing only I-frames.

( applies to: TMPGenc, HCenc, ffmpeg, Mencoder, mpeg2enc, etc. )

[edDV]

Note that I frame only MPeg2 has no temporal compression so for quality you need to use 2x-6x bitrate.

What is your source?

[Videogamer555]

Note that I frame only MPeg2 has no temporal compression so for quality you need to use 2x-6x bitrate.

Actually wouldn't it make more sense that temporal compression has lower quality so that removing the temporal compression will increase the quality without increasing the bit rate?

[jagabo]

Note that I frame only MPeg2 has no temporal compression so for quality you need to use 2x-6x bitrate.

Actually wouldn't it make more sense that temporal compression has lower quality so that removing the temporal compression will increase the quality without increasing the bit rate?

No. I-frames encode the entire picture like a JPEG image. So, for example, in a talking head shot an all i-frame video has to encode the static background as well as the moving speaker in every frame. With predicted frames it only has to encode the part of the picture that changes at each frame. Maybe just a few bits to represent the speaker's moving lips. Or in a panning shot, predicted frames can say "shift the entire image left by 4 pixels*, add fill in the right edge with this" -- much less information than encoding the entire frame. So all i-frame encoding takes a lot more bitrate. It's the whole reason predicted frames were invented.

* In practice, MPEG 2 does this on 8x8 blocks, not the entire picture.

[edDV]

Note that I frame only MPeg2 has no temporal compression so for quality you need to use 2x-6x bitrate.

Actually wouldn't it make more sense that temporal compression has lower quality so that removing the temporal compression will increase the quality without increasing the bit rate?

NO

[Videogamer555]

Note that I frame only MPeg2 has no temporal compression so for quality you need to use 2x-6x bitrate.

Actually wouldn't it make more sense that temporal compression has lower quality so that removing the temporal compression will increase the quality without increasing the bit rate?

No. I-frames encode the entire picture like a JPEG image. So, for example, in a talking head shot an all i-frame video has to encode the static background as well as the moving speaker in every frame. With predicted frames it only has to encode the part of the picture that changes at each frame. Maybe just a few bits to represent the speaker's moving lips. Or in a panning shot, predicted frames can say "shift the entire image left by 4 pixels*, add fill in the right edge with this" -- much less information than encoding the entire frame. So all i-frame encoding takes a lot more bitrate. It's the whole reason predicted frames were invented.

* In practice, MPEG 2 does this on 8x8 blocks, not the entire picture.

Thanks for the explanation. That makes more sense.

Which should I use for increased quallity, Variable Bit Rate, or a Constant Bit Rate with the constant rate set fairly high? Also will this difference between VBR and CBR effect "Iframe only" video, or is it only for when P and B frames are used?

[edDV]

What type of source are you encoding?

I-frame only is typically used to speed editing/effects computation, for example digital intermediates used in media production.

For GOP based formats, a high CBR based format is often used for acquisition (e.g. camcorders, animations). These may be edited as is (e.g. HDV, XDCAM formats), or converted to a digital intermediate. As an example, HDV is MPeg2 with 15 frame GOPS and CBR of 25 Mb/s. When converted to an I frame only digital intermediate like Cineform, the bit rate (and file size) increases 3x-5x.

After editing is finished, the final product is VBR compressed for the distribution format (e.g. DVD, Blu-Ray, TV broadcast) to maximize compression while maintaining adequate bit rate for high motion complex scenes.

[jagabo]

CBR can be easier on an editor than VBR. At very high bitrates CBR and VBR quality are similar.