So if I want to be as accurate as possible with my entry points, what method would you recommend? I have found one method to work. I load my source in TMPGEnc, and using the Manual VBR setting, I set my entry point frames to be GOP I-frames. I then divide these frame numbers by 23.976 to get the time code in seconds, which I set in my XML file. This seems to work very precisely. I have found that if I encode in anything other than Manual VBR, VCDImager complains that it can't find the entry point exactly, and sometimes places the actual entry point as much as six seconds from where I requested. So the Manual VBR seems the best option. It might work just as well if I set the I-frames to be simply I-frames (and not GOPs as well) but I haven't verified this.

- digvid