[svn] / branches / dev-api-3 / xvidcore / todo.txt Repository:
ViewVC logotype

View of /branches/dev-api-3/xvidcore/todo.txt

Parent Directory Parent Directory | Revision Log Revision Log


Revision 392 - (download) (annotate)
Wed Sep 4 18:44:41 2002 UTC (17 years, 3 months ago)
File size: 2523 byte(s)
This commit was manufactured by cvs2svn to create branch 'dev-api-3'.
todo


*** build environment ***

- portable makefile [DONE]

- QuickTime frontend
- Java frontend


*** api ***
- decore/encore2 wrappers; [DONE]
- xvid_init() [DONE]
	handle cpu detection, yuv/dct lookup table init	[DONT]
	return api version number [DONE]
	return core build number [HALTED; how?]

*** common ***

- mpeg-4 quantizer (quant4) testing [DONE]
- consder moving a modified mbprediction/mbdeprediction here [DONE]
- add [-2048,2048] to mmx dequant [DONE]
- mmx quant4 [DONE]
- sse2 [DONE]

- sse2 quant4
- mlib
	

*** image ***

- sort out in/out & conversion func naming issues [DONE]
- rgb16. [DONE]


- input flipping support
output flipping support is not 'perfect'.
using negative height is bad, best to have seperate vflip parameter

- rgb24/32_mmx cause yellow dots to appear on left hand size
(only visibile with high contrast colors)

- investigate difference between basic-yuv and iso-yuv


*** decoder ***

- quant=1 encoder material appears "soft" [FIXED]
- block based interpolation [DONE]

- add basic mpeg4 stuff:
	- custom mpeg quant matrixes [DONE]
	- intra_dc_vlc_threshold [DONE]
	- quarterpel  [in progress]
	- resync?

- msmpeg-4 support [IN PROGRESS]
- post processing [IN PROGRESS]
- b-vops [IN PROGRESS]

- proper handling of time codes and fatal/non-fatal errors


*** encoder ***

- purple mb block (rare) [FIXED]
- coeff range checks need to be verified against mpeg-4 spec [DONE]
	- dequants now clamp to [-2048,2047]. [DONE]
	- fdct_mmx clamps to [-2048,2047], it doesnt have to (but prevents overflows in mmx quantization)
	- quants _should_ clamp to [-127,127], but this introduces errors at low quant levels so we ignore it
	- idct _should_ clamp [-256,255], but its really unnessary overhead.
- pmvfast bug fixing [DONE]
- 'smearing bug' when motion occurs in straight vertical direction (
http://www.videocoding.de/forum/viewtopic.php?topic=71&forum=2&6 )
coefficient thresholding problems? [FIXED/REMOVED]
- better luminance masking [DONE]


- quater pixel motion search [IN PROGRESS]
- bidirectional coding [IN PROGRESS]
- virtualdub support [IN PROGRESS]

- global motion compensation
- redo coefficient thresholding like described in H26L
- further improve adaptive quantisation, maybe implement JND maps
- use N_VOP when a interframe is entirely composed of ublks & frame dropping. eg. simple ublk threshold


*** gpl conversion ***

mbdecoding.c [REPLACED]
encoder.c	[REPLACED]
mbmotionestcomp.c	[REPLACED]
mbcoding [REPLACED]
putvlc.c [REPLACED]
ratecontol.c [REPLACED]

No admin address has been configured
ViewVC Help
Powered by ViewVC 1.0.4