|
15117
|
1 HOW TO TEST SNOW
|
|
|
2 ----------------
|
|
|
3
|
|
|
4 Snow is an experimental wavelet-based codec made by the FFmpeg developers,
|
|
|
5 and while it is still in heavy development, it is already giving very good
|
|
|
6 results.
|
|
|
7 Be very careful though, as the format of the bitstream produced might
|
|
|
8 change, do not rely on it to store videos that you value.
|
|
|
9 For this reason, MEncoder will not encode without 'vstrict=-1' on the
|
|
|
10 command line.
|
|
|
11
|
|
|
12
|
|
15122
|
13 OPTIONS RECOGNIZED BY SNOW
|
|
15117
|
14
|
|
|
15 * vqscale=<0.01-255>
|
|
15122
|
16 Encoding quality, sane range 1-10, default: 0 (lossless). May be
|
|
|
17 fractional.
|
|
15117
|
18 A given quality in snow needs a somewhat lower qscale than the same
|
|
|
19 quality in MPEG-4.
|
|
|
20 Note that 0 may not be specified; if you want lossless encoding, you
|
|
15122
|
21 must leave out vqscale.
|
|
15117
|
22
|
|
|
23 * cmp, subcmp, mbcmp
|
|
|
24 Set the comparison function, default: 0 (SAD).
|
|
|
25 useful values = 0 (SAD), 1 (SSD), 11 (5/3 wavelet), 12 (9/7 wavelet).
|
|
|
26 Experience shows that SSD is the best most of the time, while SAD is
|
|
|
27 slightly better the remainder of the time.
|
|
|
28 You can add 256 to any of the options to enable chroma motion
|
|
|
29 estimation for that comparison (e.g. mbcmp=257 for SSD with chroma),
|
|
|
30 but it doesn't seem to help much for the moment.
|
|
|
31
|
|
|
32 * pred=<0-2>
|
|
|
33 Wavelet type. 0 (9/7 wavelet), 1 (5/3 wavelet), 2 (13/7 wavelet).
|
|
|
34 9/7 is probably better for for lossy coding, and 5/3 for lossless.
|
|
|
35
|
|
|
36 * last_pred=<0-3>
|
|
|
37 Tries a few extra predicted motion vectors before doing EPZS search,
|
|
15122
|
38 default: 0.
|
|
15117
|
39 This option has negligible effect on both speed and quality of snow,
|
|
15122
|
40 so just leave it off. (It does, however, help MPEG-4.)
|
|
15117
|
41
|
|
|
42 * qpel
|
|
|
43 Refines motion estimation, default: off.
|
|
|
44 This setting always helps compressibility, but costs some CPU time
|
|
|
45 both while encoding and decoding.
|
|
|
46
|
|
|
47 * v4mv
|
|
|
48 Allows smaller motion partitions, default: off.
|
|
|
49 The current MB decision algorithm doesn't make very good use of this:
|
|
15122
|
50 It improves quality, but also increases bitrate. (You could get
|
|
15117
|
51 more quality per bitrate by reducing quantizer instead.)
|
|
|
52
|
|
|
53
|
|
|
54 v4mv and the wavelet cmp functions are theoretically good, but in practice
|
|
|
55 won't work well until we get OBMC-aware motion estimation.
|
|
|
56
|
|
|
57 In short:
|
|
|
58 The best options in almost all cases are
|
|
|
59 vcodec=snow:vstrict=-1:vqscale=$N:pred=0:cmp=1:subcmp=1:mbcmp=1:qpel |
