TM5 Overview

The Test Model served as a cook book for creating bitstreams during the collobrative co-exerpimental phase of MPEG-2 video. The many documented experiments were an attempt to verify the usefulness of various proposed coding techniques. Each participant would follow the recipe given in the TM document to create an encoder. If the proposal met the criteria (coding gain, implementation complexity, robustness)--it survived. In the end, only a handful made their way into the MPEG syntax.

The last major update of the Test Model document, version 5, took place at the Sydney, Australia meeting of the MPEG working group (WG11) in March 1993. Since the MPEG-2 main profile syntax froze at this meeting, only limited scalability experiments continued past March 1993. A small two page delta document describing Temporal Scalability experiments, affectionately called "TM 6", was produced at the New York City meeting in July 1993. From then on, direct bitstream exchanges among participants of the working group helped to resolve ambiguities in the official MPEG working draft document which ultimately became the official standard document (ISO/IEC 13818-2, ITU-T H.262) we all know and love today.

History:

The Test Model evolved in parallel with the MPEG video working draft. The TM series was a joint effort between ITU-T SG15.1 (known then as CCITT SG XV, Working Party XV/1, Experts Group on ATM Video Coding) and ISO/IEC JTC1/SC29 WG11 (MPEG).

 Version  Meeting location  Date     MPEG doc. no.   ITU-T SG15 doc no.
 -------  ----------------  ------   -------------   ------------------
 TM 0     Singapore         Jan 92   [ known as Video Working Draft 0 (to keep Cesar happy) ]
 TM 1     Haifa, Israel     Mar 92   MPEG 92/160     AVC-260
 TM 2     Rio, Brazil       Jul 92   MPEG 92/245     AVC-323
 TM 2.2   Tarrytown, NY     Oct 92   MPEG 92/535     AVC-356
 TM 3     London            Dec 92   WG11 92/328     AVC-400
 TM 4     Rome              Feb 93   MPEG 93/225b    AVC-445b
 TM 5     Sydney            Apr 93   MPEG 93/457     AVC-491
 TM 5b    Sydney            Apr 93   WG11 93/400     AVC-491b
 "TM 6"   New York City     Jul 93   -               -

See Leonardo's nostalgic meeting record for a complete list of meetings.

Parent models:

The first drafts of the coding models inherited much of the coding methods and documentation style from previous standardization efforts.

 MPEG-1: final document was SM-3 (Simulation Model version 3)
 H.261:  final document was RM-8 (Reference Model version 8)

TM 5 Main Profile relevant sections

Sections 1-8, 10, Appendix B, C, K and L of Test Model 5.0 apply to the Main Profile syntax of today. Of course, corresponding sections in the official MPEG document (IS 13818-2 / ITU-T H.262) always supersede those in the Test Model. Section 9, Appendix D through L are mostly scalability sections, many of which no longer even apply to the scalability syntax of today.

The following is a list of comments on those sections which DO apply to Main Profile:

1 Introduction

The document is no longer maintained. Please do not contact the editor (Koster) unless for historical purposes :-)

2.3.2 Hierarchical profile

- features in "Hierarchical profile" merged into High Profile. - "Harmonized scalable solution" was never found. SNR and Spatial Scalable streams remain separate today.

2.3.3 Professional profile

- features in "professional profile" merged into High Profile.

3.3.1 conversion of CCIR 601 to the 4:2:0 format

Although chroma subsampling is a display process which falls outside the official scope (normative) of the MPEG standard, it should be done properly for quality reasons. The Encoder's chroma decimator should anticipate the behaviour of the Decoder's chroma interpolator. The MPEG bitstream header element chroma_420_type was meant to convey which of two methods (interlaced or progressive) a decoder should use to upsample (interpolate) the decoded chroma output. By November 1994, chroma_420_type became synonymous (and therefore redundant) with the progressive_frame header element. chroma_420_type was kept in the syntax for compatibility reasons. After all, the syntax was frozen over 1 year earlier!

Test Model 5 gives two sinc conversion filters suitable for intra-field vertical chroma subsampling in interlaced video:

    {-29,0,88,138,88,0,-29}/256  for field 1
    {1,7,7,1}/16                 for field 2
 
      4:2:2   4:2:0

        Odd

        Even    X

        Odd

        Even    X

        Odd

        Even    X
  
        Odd

  An inter-field (intra-frame) filter would be used for progressive 
  video sequences which renders the subsampled chroma components centered
  between the spatial positions of the original components.

      4:2:2   4:2:0

        Odd
                X
        Even     

        Odd
                X 
        Even     

        Odd
                X
        Even     
                 

The progressive vs. interlaced vertical chroma filter was later resolved at the November 1993 meeting in Seoul, Korea. Look at section 6.1.1.8 of the final MPEG-2 video document (IS 13818-2 / ITU-T H.262) for a description of interlaced and progressive chroma sample siting.

4. Layered Structure of video data

The official MPEG video document (IS 13818-2 section 6) does an excellent job of describing the layred structure of video data.

5. Motion Estimation and Compensation

See the modified Chapter 5.

6. Modes and Selection

See the modified Chapter 6.

7. Transformation and Quantization

See the modified Chapter 7.

8. Coding

See the modified Chapter 8.

- An additional informative example of encoding is described in Appendix D of the MPEG-1 video document (IS 11172-2).


The following comments point out irrelevant sections with respect to today's Scalable MPEG-2 syntax:

9. Video multiplex coder

Section 9 never discusses general or Main Profile syntax.

- frequency scalability was dropped from the MPEG toolkit.

- the syntax and semantics for scalability in the MPEG-2 video document supersede those present in this section of TM 5. The are quite different today than they were in April 1993.

10. Rate Control

See the modified Chapter 10.

Appendix A: discrete cosine transform

See the modified Appendix A.

Appendix B: variable length code tables

This section is extremely redundant with Appendix B of IS 13818-2. All the official VLC tables are listed there.

Appendix C: video buffer verifier

The TM only referred to the video specification Working Draft... and rightly so!

Appendix D: frequency domain scalability extension

- again, frequency scalability was dropped from the MPEG toolkit.

Appendix F: cell loss experiments

- this experiment could be generic enough and mostly independent of scalability to apply to Main Profile, although it does refer to jettisoned items such as leaky prediction.

F.4 AC-Leaky Prediction

- leaky prediction was dropped from the MPEG toolkit.

F.5 Data partitioning vs. 1-layer cell loss experiment

- data partitioning is still a part of the MPEG syntax (toolkit), but is yet not enabled in any MPEG Profile.

Appendix G: Compatibility and Spatial Scalability

- the weights for spatial scalablity were modified several times before the final November 1994 video standard draft.

- the only items from Appendix G which would still apply to the scalability syntax are those high-level parameters which specify bitrate, M and N factors, picture types, etc.

Appendix H: Low Delay Coding

- low delay coding is still very much a part of the Main Profile syntax, though is not commonly used. See section D.5 in the MPEG-2 video document for a description.

H.1.6 Influence of leaky prediction on low delay coding

- again, leaky prediction is not a part of the MPEG toolkit.

Appendix I: Frequency Domain Scalability Core Experiments

- again, frequency scalability was dropped from the MPEG toolkit.

Appendix J: Harmonized Hybrid Scalability

- again, no "Unified Field Coding" for scalability was reached. Maybe MPEG-4 will do this. (supressing urge to comment)

Appendix K: Fast Forward and Fast Reverse Modes

- intra_slice is still a legitimate trick mode technique (and about the only one viable today).

K.4 data partitioning approach

- again, data partition is still a part of the MPEG toolkit (cf section 7.10 in IS 13818-2 / ITU-T H.262), but yet not enabled by any Profile and/or Level combination. This section does however demonstrate one of Data Partitioning's primary intended uses.

Appendix L: Data partitioning

- see comments above

- syntax for data partitioning has changed. In particular, there is no data_partitioning_flag, intra_pbp in the sequence and picture headers, respectively. Instead, Data partitioning mode would be signaled by the scalable_mode element (if a tag were defined for it). The pbp element in Appendix L of the TM 5 document is now called priority_breakpoint in the MPEG-2 video document.

Appendix Q: Quantization

- The 8x1 DCT mode was an attempt to solve the ringing artifacts present around "text" and other sharp edges within reconstructed pictures. It is not a part of the MPEG toolkit. Careful coding using existing syntax tools (macroblock quantizer_scale) is recommended.


  • Back to Contents