Should I set "B" to I-frame when I split a video?

[kccmo]

I should set "B" to I-frame?
or should set "B" to previous frame of I-frame?

I-frame is a key frame, it is a complete image, following frames will refer to this frame, so the "A" must be I-frame.
But how about the "B"?

"B" will be the last frame of video after split, no frame will refer to it, is it unnecessary to set "B" to I-frame?

If I should set "B" to I-frame? How about this case?

I split a video to two, for some reasons, I want join them to one again.

If I set I-frame to last frame of first video, will it duplicate with first frame of second video?

[usually_quiet]

It depends on the software you use for splitting the file and what you want to do with the file

As I recall, if the file isn't going to be played until it is joined again, you could use 7z/Zip/Rar with file-spanning or HJSplit and split the file on any frame.

If the pieces of the file must be playable, typically you would use an editor of some kind. To ensure the entire clip is playable, editors typically don't allow dividing a file into pieces that don't start with an I-frame and end with either a P-frame or an I-frame. If a clip doesn't start and end with the right type of frame, many editors will re-encode all or part of the clip so it does start and end appropriately. Editors that don't re-encode probably won't let you start or end on the wrong kind of frame, so to get the entire file, there will be overlap at I-frames. When joined again, the I frames at the joins would be duplicates.

[pandy]

Depends from GOP structure, generally you can cut video at independently accessible video frames in most of cases this mean that you need I frame.
Side to this you can't mark I frame as B or B as I - you must reencode video frame as I frame. And combining to video (concatenate) will give you only one I frame (as preceding P frame usually belongs to preceding I frame - it is part of different GOP - valid for closed GOP, for open GOP things are more complicated).
Some encoders allow to place I frame at the end - this mean that concatenating with second video you will have I I - this may violate your max bitrate constraints.

[kccmo]

p.s.

I am talking about split video without re-encoding, e.g. avidemux, MKVToolNix

[pandy]

p.s.

I am talking about split video without re-encoding, e.g. avidemux, MKVToolNix

This was my impression... You can't set B frame as I - B frame refer to previous I and P frames and usually it is very small (usually B frame cary lest than 1 - 10% of data stored in I frame) - if you try to recreate picture from B frame data, without referring to other I and P you will receive green screen with only few macroblocks (idealized test).
There is group of smart video encoders that allow to edit video recoding only what need to be recoded if cut point is not at I boundary.
They can split GOP on non I frame and recode only this GOP to new one - remain video are left not touched.

http://wiki.multimedia.cx/index.php?title=Frame_Types

[usually_quiet]

p.s.

I am talking about split video without re-encoding, e.g. avidemux, MKVToolNix

Assuming the the GOPs follow a pattern something like IBBPBBPBBPBBIBBPBBPBBPBBI, you must start and end each section with an I frame to keep all the video and avoid re-encoding.

If you split my sample above into two sections, the result would be IBBPBBPBBPBBI in one section and IBBPBBPBBPBBI in the other.

When the sections were joined again using an editor you would have IBBPBBPBBPBBIIBBPBBPBBPBBI

Most people would not notice the repeated I-frames if the sections are large and few in number. If the file is cut into many tiny sections more people would notice.