[XviD-devel] todo/action list

[Christoph Lampert]

On Tue, 3 Sep 2002, peter ross wrote:

> >I hope everyone knows the debugview output the VFW generates.
> >
> >Why not putting exactly that stuff into a file? Would be easy to do, and 
> >it's easily parsable.
> 
> some comments:
> 
> * we should not store the field name on each line of the file.
>   it wastes space, and makes it harder to parse (right gruel?)

Right. There could be a "comment" in the first line naming the fields,
best would be to let it start with a '#', that's comment sign in 
gnuplot and many others ;-)

> * what values do we need?
>   - frame_size   (bytes)
>   - header_bits  (VOL/VOP)
>   - texture_bits
>   - quant_type
>   - quant
>   - kblocks
>   - mblocks
>   - ublocks

frame_type I, P or B has to be saved! 

Maybe also flags as hex or oct? I could get interesting when there is
dynamic switch of feature. Or will this always stay disabled? quant_type
would be included, then. 
The _read_ routine should btw. _not_ rely on an absolute number of fields,
but be compatible with "extension" to the right. 

I would like something about "absolute motion vector length", too. 
Because there isn't much difference between no motion and 
fast, but simple motion, in terms of sizes. If we do statistics
anyway, "absolute motion variance" might be interesting, too, although
this is closer to motion_bits. 

One day, segmentation information and/or GMC data could be saved as well,
(segmentation in binary form).

gruel