NAME

mpeg_stat - analyzes MPEG-1 encoded bitstreams

SYNOPSIS

mpeg_stat [ -all basename ] [ -block_info filename ] [ -dct ] [ -end N ] [ -histogram filename ] [ -offsets filename ] [ -qscale filename ] [ -quiet ] [ -rate filename ] [ -ratelength N ] [ -time ] [ -size filename ] [ -start N ] [ -syslog filename ] [ -userdata filename ] [ -verify ] [ file_name ]

DESCRIPTION

mpeg_stat decodes mpeg-1 encoded bitstreams collecting varying amounts of statistics. Errors are sent to stderr, basic statistics to stdout. The basic information is the pattern of frames used, number of bytes for each frame type, the specified parameters, and lengths of vectors. For each frame type, the average size, compression rate, Q-factor, and time to decode are given. Wherever a filename is requested - can be used to denote stdin/out.

OPTIONS

-all basename : records information from all options (qscale, block_info, etc.) into files with basenames of basename (For wizards, -all foo -block_info bar will record all options BUT -block_info into files of the form foo.*).

-block_info filename : records information about every block into file filename. Useful for detailed analysis and to make specifics files for re-encoding (see block2spec). The format is:

frame frame# IPB-type mv-scale temporal_ref

slice slice# q-scale

block block# IPB-type Q-scale bits block-type <vectors-if-any> <cbp> <dct decode>

Numbering starts at 0 (except slices) and all motion vectors are in half-pel units <X Y>.

-dct : enables the decoded dct values to be output into the block file.

-end N : causes mpeg_stat to finish collecting statistics at frame N.

-histogram filename : summaries the overall statisitcs for each frame type into the file.

-offsets filename : record the offset of every picture, GOP, and slice into file filename.

-quiet : toggles the display of the frame-types as they are parsed (and custom quantization matrices) [default: on].

-qscale filename : writes Q factor and custom quantization matrix information into file filename. The Q-factor and number of blocks of that quality (ignoring skipped blocks) for each type of frame is recorded.

-rate filename : record the bit rate at every picture (after the first second) into the file filename. Also collect minimum and maximum bit rate encountered in the summary information. (Rate is the number of bits used per second, at every frame).

-ratelength N : Change the measurement period for rate to N, so it will measure the total rate required over any N frame period (after the first N, of course).

-size filename : stores information about the type (I,P,B) and size of each frame in file filename (one frame per line).

-start N : causes mpeg_stat to begin collecting statistics at frame N. (Preceding frames will be parsed, but few statistics are collected (the system layer counters will sill be running).) -syslog filename : describes the parsing of the system layer into the file filename. Note this option is not turned on by -all. -time : prints the amount of time total and per-frame type averages to decode. This will give an estimate of time in a software player, but is not generally useful, so it is off by default. -userdata filename : dumps user data fields into a file. Can be read in ASCII. -verify : does more work to check the validity of the sequence. Slows down the statistics, so it is optional.

EXAMPLE

Decoding the flowergarden sequence (mpeg_stat -quiet flower.mpg) should produce the following statistics:

Reading /u/smoot/mpg/flower.mpg

Frame sequence as to be displayed: BBPBBPBBPBBPBBIBBPBBPBBPBBPBBIBBPBBPBBPBBPBBIBBPBBPBBPBBPBBIBBPBBPBBPBBPBBI BBPBBPBBPBBPBBIBBPBBPBBPBBPBBIBBPBBPBBPBBPBBIBBPBBPBBPBBPBBIBBPBBPBBPBBPBB

Searching for constant frame type sequence...pattern detected: IBBPBBPBBPBBPBB

SUMMARY:

Total Bytes read: 0. Total number of frames: 150. Length is 5.01 sec

Width: 352 Height: 240 Avg. Frame Size: 4786 bytes + 6 bits (average rate 1147692.76 bits/sec)

Total Compression Rate: 1.89 % of uncompressed 24 bit images = 0.45 bits per pixel

Number of Macroblocks [width * height = sum]: 22 x 15 = 330 per frame Skipped Macroblocks = 4389 (9.46%), Coded Macroblocks = 41992 (90.54%)

Total Time Decoding: 8.611 secs. 0.05721 sec/frame or 17.11 frames/sec.

MPEG-Viewer requirements: Pixel aspect ratio of 0.6735 Required display speed: 29.97 frames/sec Specified bit rate is 1.00 MBits/sec (2500 * 400bits/sec) Requested buffer size is 8K ints (16 bits). And the constrained parameter flag is on. The stream meets the constrained parameter requirements.

Length of vectors in pixels: Horizontal forward vectors, maximum : 39 average: 6 Vertical forward vectors, maximum : 22 average: 1

Horizontal backward vectors, maximum: 29 average: 5 Vertical backward vectors, maximum : 16 average: 1

Frame specific information:

10 I FRAMES, average is: Size: 17323 bytes + 1 bits (24.13%) Compression Rate: 6.84% Q Factor [scales quantization matrix]: 11.15 Time to Decode: 0.114039 secs.

40 P FRAMES, average is: Size: 8031 bytes + 0 bits (44.74%) Compression Rate: 3.17% Q Factor [scales quantization matrix]: 10.85 Time to Decode: 0.075342 secs.

100 B FRAMES, average is: Size: 2235 bytes + 4 bits (31.13%) Compression Rate: 0.88% Q Factor [scales quantization matrix]: 15.01 50.59% interpolated Macro Blocks Time to Decode: 0.044516 secs.

NOTES

The analyzer expects MPEG-1 video streams only. It can handle multiplexed MPEG streams (video+audio streams), but mostly analyzes the video portion.

Some streams do not end with the proper sequence end code and will probably generate an "Improper sequence end code." error when done. In general mpeg_stat attempts to catch and alert its user to errors in the stream. Such errors are glossed over in mpeg_play. Be aware that errors can disturb statistics gathering, generating odd results.

This player can handle XING data files. Be aware that XING makes no use of temporal redundancy or motion vector information. In other words, they do not use any P or B frames in their streams. Instead, XING data is simply a sequence of I frames.

HISTORY

The analyzer is based on the UC Berkeley mpeg_play player by Ketan Patel, Brian Smith, Henry Chi-To Ma, and Kim Man Liu. It was modified at the Technical University of Berlin, Germany, Dept. of Computer Science by Tom Pfeifer, Jens Brettin, Harald Masche, Alexander Schulze, and Dirk Schubert. It has been further modified to collect much more information by Steve Smoot (UC Berkeley).

BUGS

No statistics should be collected before the start frame. Sometimes system layer streams will indicate that they lack proper end codes, when they have them. VBV size conformance is not checked. Verification is not complete. Frame numbers should probably be in display numbering not stream numbering. Other bugs? Send mail to <mpeg-bugs@plateau.cs.berkeley.edu>.

VERSION

This is version 2.2, contining some new features since 2.1, and several bug fixes. It is a major change since version 1.0.

AUTHORS

Ketan Patel - University of California, Berkeley, kpatel@cs.berkeley.edu

Brian Smith - University of California, Berkeley, bsmith@cs.berkeley.edu

Henry Chi-To Ma - University of California, Berkeley, cma@cs.berkeley.edu

Kim Man Liu - University of California, Berkeley, kliu@cs.berkeley.edu

Tom Pfeifer - Multimedia systems project - pfeifer@fokus.gmd.de

Steve Smoot - University of California, Berkeley, smoot@cs.berkeley.edu