DSV

DSV (Digital Subband Video 1, or DSV1) is a hobby codec by EMMIR (Envel Graphics): a small, wavelet-based, software-only codec that the author compares to MPEG-1/MPEG-2 in efficiency. It exists to be read and tinkered with, not to play back in anything. The repository was archived on 4 October 2024 and its README points to DSV2, which the author says compresses far better. Use DSV2 for any new experiment.

What it does

• Multiresolution subband (wavelet) transform instead of a DCT.
• Intra and forward-predicted inter frames, half-pixel motion compensation, no B-frames, variable-length closed GOPs.
• 8-bit 4:1:1, 4:2:0, 4:2:2 and 4:4:4; no interlacing; frame dimensions must be even.
• Interleaved exp-Golomb entropy coding, adaptive quantisation, and hierarchical motion estimation.
• Single-pass ABR or CRF rate control.
• The reference implementation is C89 with integer-only maths, no dependencies and a single thread.

Using the reference tool

The author’s reference code (also available as the single header dsv1.h) is the only implementation; there is no FFmpeg or player support. Build it with cc -O3 -o dsv1 *.c from a checkout and run:

./dsv1 e -inp_video.yuv -out_compressed.dsv -w352 -h288 -fps_num24 -fps_den1 -qp85 -gop15
./dsv1 d -inp_compressed.dsv -out_decompressed.yuv -out420p

Options are written as -name immediately followed by the value (-w352, -qp85), unlike DSV2’s -name=value. Input is raw YUV, so give the dimensions and frame rate; it is read as 4:2:0 unless you set -fmt. -kbps switches to ABR. -out420p converts the decoded output to 4:2:0, which discards chroma from a 4:2:2 or 4:4:4 stream. The format is described in DSV1_spec_r2.pdf in the repository.