| 1 |
IN PROGRESS |
| 2 |
|
| 3 |
- Remove obsolete i,j in macroblocks in favour of Coord.... |
| 4 |
- Replace appending subsampled luminance to luminance data with a 5-plane |
| 5 |
picture data model (YUVHQ) |
| 6 |
- MotionEst should be called 'MacroBlockCodingMode' |
| 7 |
|
| 8 |
- Set satlim and scale type in quantizer during init |
| 9 |
|
| 10 |
- Move to a buffering output model where a frame can be aborted/restarted |
| 11 |
with different quantisation if it is coming out too large or too small. |
| 12 |
|
| 13 |
- Dynamic B/P selection and selection of I frames |
| 14 |
without memory-expensive look-ahead. |
| 15 |
|
| 16 |
- Improve memory management. Selecting Picture's from a single pool |
| 17 |
rather than seperate B / I|P pools will save memory buffers. Also, |
| 18 |
really there is no need for DCT blocks for every picture as encoding |
| 19 |
is basically serial!! |
| 20 |
|
| 21 |
|
| 22 |
Near Future improvements |
| 23 |
|
| 24 |
0. Replace current variance based motion estimation mode selection |
| 25 |
with DCT based selection. |
| 26 |
Next Step: split predictors from Picture state and build a |
| 27 |
new QuantiseAndPutEncoding that predict's and transforms and |
| 28 |
quantisatises forall the motion estimates and selects the one that |
| 29 |
worked best for encoding. |
| 30 |
|
| 31 |
- Modify the on-the-fly rate controller so it can handle adjusting |
| 32 |
quantisation at select Macroblock's and not at *every* macroblock |
| 33 |
if no fancy per-block quantisation fiddles are active... |
| 34 |
|
| 35 |
1. Add single coefficient dropping to DCT... |
| 36 |
|
| 37 |
2. Split into front-end and library (for gstreamer etc) |
| 38 |
This will require removal of lots of nasty global scratch variables |
| 39 |
|
| 40 |
3. Add accurate (GOP-structure based) decoder buffer modelling for more |
| 41 |
aggresive bit-rate optimisation. |
| 42 |
|
| 43 |
4. Seperate the transformation phase (ME, pred, DCT) from the bit allocation |
| 44 |
and coding phases. |
| 45 |
This will provide the basis for two-pass encoding and pipelined encoding. |
| 46 |
|
| 47 |
|
| 48 |
5. The rate control code is now a mess of int64_t , double and int. Needs |
| 49 |
tidying... |
| 50 |
|
| 51 |
6. Add support for the ATSC Active Format Description and Bar Data extenstions |
| 52 |
|
| 53 |
|
| 54 |
Minor Tidying |
| 55 |
|
| 56 |
1. A lot of parameter passing could be removed from motionest.cc routines |
| 57 |
by creating an 'MacroBlockMotionModes' object carrying parameters fixed |
| 58 |
for a block... |
| 59 |
|
| 60 |
2. The order of return from block_dist1_MMXE is silly. Why not |
| 61 |
retune in the order it is held and adjust find_best_one_pel_MMXE to |
| 62 |
suit. |
| 63 |
|
| 64 |
3. For Motion estimation in -I 1 we really need both field *and* frame |
| 65 |
sub-sampled images for motion estimation to be optimal. |
| 66 |
|
| 67 |
|
| 68 |
5. The code for scene-change gop sizing and ending sequences can in |
| 69 |
some circumstances mess up as the sizing takes no account of |
| 70 |
the fact the when a sequence split occurs the I-frame tempref is |
| 71 |
0 not M-1. |
| 72 |
|
| 73 |
|
| 74 |
Sometime.... |
| 75 |
|
| 76 |
|
| 77 |
|
| 78 |
1. Bring field sequence MC/prediction to same level as progressive |
| 79 |
sequence. |
| 80 |
|
| 81 |
|
| 82 |
2. Improve locality of reference by motion-estimating entire 4x4, 2x2, 1x1 |
| 83 |
phases in a single pass rather doing each for each Macroblock. |
| 84 |
|
| 85 |
3. Check out using dist2 for initial MC calculations as well as final |
| 86 |
selection? |
| 87 |
|
| 88 |
|
| 89 |
5. Add low-passing for high-activity regions when a specified quantisation |
| 90 |
level is exceeded (or block boundary edge energy exceeds some threshold?). |
| 91 |
|
| 92 |
6. Sort out that wretched vbv verification code. |
| 93 |
|
| 94 |
The mblockq_ sub-routines intended specification should be documented... |
| 95 |
|
| 96 |
|
| 97 |
|
| 98 |
|