Bad I-frames on QuEnc in MPEG1

[Ark]

I've done a small clip for test the various encoding options of QuEnc (v0.51), but the stream is somewhat bad.

Loading the resulting mpeg1 in vdub showed that every Keyframe seems to be encoded at a much higher quantizer that P-B frames, causing very noticeable blocking artifacts everywhere.

I've tried every possible combination of options, but the issue remains, but only in mpeg1 encodings. Mpeg2 is flawless.

The source is a small .avs with some trims from the dvd "Princess Mononoke", 1 minute long, 352x288 pixels without filters, only resizing.

I hope that this can help on improving this very cool program!

[Mug Funky]

VBR or CBR? i'd imagine with CBR the i-frames would need a higher quant to be in line with its much smaller neighbors.

ffmpeg's rate control isn't it's strongest feature though. it's definitely better than when nic found it :)

[Ark]

Originally posted by Mug Funky
VBR or CBR? i'd imagine with CBR the i-frames would need a higher quant to be in line with its much smaller neighbors.

ffmpeg's rate control isn't it's strongest feature though. it's definitely better than when nic found it :)

Err...both :D

It seems to be a mpeg1-only related bug, regardless of encoding options chosen...

It creates a curious "fluctuating" effect so that in every GOP the I frames gets very blocky but it's very hardly noticeable in some scenes, while becoming clearly visible in others...

I wanted to make some VCDs but for now...

So, I'll patiently(is this word correct?) wait for an improvement of rate control!

[dragongodz]

actually it does happen with mpeg2 aswell just not so bad or noticable.

and yes it is the libavcodec rate control thats FUBAR. try mencoder and ffmpeg and you will get the exact same problem.

[CavalloPazzo]

On some test I did with mencoder some time ago I noted that with trellis quant. enabled this was more visible. Anway turning trellis off and increasing i-bias solved the problem for me