Dear MPEG colleagues, tcela-11v2 is an MPEG-1 constrained bitstream that demonstrates that some perfectly legal and compliant bitstreams cannot be decoded properly by some perfectly compliant decoders. IMPORTANT: This bitstream is NOT intended to used for Video Conformance Testing. However the information contained here may be relevant to the work of this AHG. The problem demonstrated here is that the domain for which an IEEE compliant IDCT is required to give correct results is not sufficient to decode all compliant MPEG (MPEG-1 or MPEG-2) bitstreams. Several steps that can occur in an encoder can cause the DCT coefficients reconstructed by the decoder (after inverse quantization) to be out of this domain: Inaccuracy of the FDCT, quantization of coefficients, coefficient clamping. This problem is addressed by one of the Video Technical Comments of the USNB. This problem exists both in MPEG-1 and MPEG-2. This example is an extreme case, in which the first P-frame was coded using a very inaccurate FDCT. According to the MPEG-1 IS, this bitstream is perfectly compliant. You should be able to match the reconstructed frames perfectly if you use a software decoder implementation of IDCT that never overflows (e.g. floating point). However, the images reconstructed by a hardware decoder may have a very large mismatch (in the upper part of the frames). Check with your hardware decoder, if you have one. Note: Not all software MPEG-1 decoder will decode this bitstream correctly (for example, the "mpeg_play" MPEG-1 X viewer from Berkeley University does not). Bit rate: 1.0 Mbit/sec Sampling dimensions: 352 x 240 x 29.97 Hz Bitstream file tcela-11v2.bits Reconstructed frames* tcela-11v2.decoded Decoding trace ** tcela-11v2.trace picture structure I P P P P... *frame format: 4:2:0 frame ("display" order) 8 bit width +---------------+ | | | Y0 | height |width/2 | +-------+-------+ | Cb0 | height/2 +-------+ | Cr0 | height/2 +-------+-------+ | | | Y1 | | | +-------+-------+ | | etc... ** Only the first row of MBs is traced. More complete traces are available on special order. Best regards, Tristan ------------------------------------------------------------- Tristan Savatier E-mail: tristan@la.tce.com THOMSON CONSUMER ELECTRONICS TEL: +1 310 568 9003 Los Angeles Laboratory FAX: +1 310 568 9002 5601 W. Slauson Ave., Suite 262, Culver City, CA 90230, USA -------------------------------------------------------------