Digital-Subband-Video-2

September 12, 2025 · View on GitHub

As of June 20, 2025 the DSV2 bitstream is frozen (v2.8)

DSV2 is a lossy/lossless video codec using wavelets and block-based motion compensation. It performs best at medium-low to medium-high bitrates at resolutions between ~352x288 (CIF) to 1920x1080 (FHD). Comparable to MPEG-4 Part 2 and Part 10 (H.264) (P frames only) in terms of efficiency and quality.

Example comparison (More can be found at the bottom of the README):

Stefan CIF minih264:

https://github.com/user-attachments/assets/e5695fa1-afa7-4205-9901-1993569efd88

DSV2:

https://github.com/user-attachments/assets/379beb83-b0ff-4349-a93b-a0b16b72e85c

Note PDFs are out of date. They reflect the codec as it was when it was first uploaded here.

DSV2 Features (refer to PDF in repo for more detail)

  • compression using multiresolution subband analysis instead of DCT
    • also known as a wavelet transform
  • up to quarter-pixel motion compensation
  • 4:1:0, 4:1:1, 4:2:0, 4:2:2 (+ UYVY) and 4:4:4 chroma subsampling formats
  • adaptive quantization
  • in-loop filtering
  • lossless coding support
  • intra and inter frames with variable length closed GOP
    • no bidirectional prediction (also known as B-frames). Only forward prediction with previous picture as reference

Improvements and New Features since DSV1

  • in-loop filtering after motion compensation
  • more adaptive quantization potential
  • skip blocks for temporal stability
  • new subband filters + support for adaptive subband filtering
  • better motion compensation through Expanded Prediction Range Mode (EPRM)
  • quarter pixel compensation
  • lossless coding support
  • psychovisual optimizations in the codec and encoder design

--- for more detailed information please refer to the informal specification document (DSV2_spec.pdf) in the repository.

Encoder Features (specific to this implementation of the DSV2 spec)

  • single pass average bitrate (ABR) or constant rate factor (CRF) rate control
  • more advanced Human Visual System (HVS) based intra block mode determination
  • new Human Visual System (HVS) based intra frame adaptive quantization
  • more complex scene change detection
  • hierarchical motion estimation
  • better temporal adaptive quantization
  • written to be compatible with C89

--- for more detailed information please refer to the encoder information document (DSV2_encoder.pdf) in the repository.

Limitations

  • no built-in interlacing support
  • only 8 bits of depth per component supported
  • frame sizes must be divisible by two

This code follows my self-imposed restrictions:

  1. Everything must be done in software, no explicit usage of hardware acceleration.
  2. No floating point types or literals, everything must be integer only.
  3. No 3rd party libraries, only C standard library and OS libraries for window, input, etc.
  4. No languages used besides C.
  5. No compiler specific features and no SIMD.
  6. Single threaded.

Compiling

C Compiler

All you need is a C compiler.

In the root directory of the project (with all the .h and .c files):

cc -O3 -o dsv2 *.c

Zig build system

If you have Zig installed, you can use the build.zig file to compile the project. Building requires Zig version ≥0.13.0.

  1. Ensure you have Zig & git installed

  2. Clone this repo & enter the cloned directory:

git clone https://github.com/LMP88959/Digital-Subband-Video-2
cd Digital-Subband-Video-2
  1. Build the binary with Zig:
zig build

Note: If you'd like to specify a different build target from your host OS/architecture, simply supply the target flag. Example: zig build -Dtarget=x86_64-linux-gnu

  1. Find the build binary in zig-out/bin. You can install it like so:
sudo cp zig-out/bin/dsv2 /usr/local/bin

Now, you should be all set to use the compiled dsv2 binary.

Running Encoder

Sample usage:

./dsv2 e -inp=video.y4m -out=compressed.dsv -y4m=1 -qp=60 -gop=48

Running Decoder

Sample usage:

./dsv2 d -inp=video.dsv -out=decompressed.y4m -y4m=1 -out420p=1

NOTE: if -inp= and -out= are not specified, it will default to standard in / out (stdin/stdout). Only .yuv (one file containing all the frames) and .y4m files are supported as inputs to the encoder.

Notes

This codec is by no means fully optimized, so there is a lot of room for performance gains. It performs quite well for what it is though.

Wavelet Codec Testbench

Check out this open source / work in progress wavelet codec testbench by Gianni Rosato!
https://github.com/gianni-rosato/wavelet-bench

If you have any questions feel free to leave a comment on YouTube OR join the King's Crook Discord server :)

YouTube: https://www.youtube.com/@EMMIR_KC/videos

Discord: https://discord.gg/hdYctSmyQJ

itch.io: https://kingscrook.itch.io/kings-crook

Example videos:

The H.264 file sizes were within a few kilobytes of their respective DSV2 file size.
H.264 examples were encoded using https://github.com/lieff/minih264 using -speed0 (best quality) unless otherwise stated
DSV2 examples were encoded with -effort=10 (best quality)

Husky CIF - 29.97fps, GOP = 60 minih264:

https://github.com/user-attachments/assets/4ae22745-0d74-4a37-96b8-dc427dce720b

DSV2:

https://github.com/user-attachments/assets/b0975e1d-74f9-4322-bbeb-0edbd81bba39

Mobile CIF - 29.97fps, GOP = 60 minih264:

https://github.com/user-attachments/assets/5562d96d-092c-46da-b0ab-1063bd0587a6

DSV2:

https://github.com/user-attachments/assets/fa2861d8-4e50-47fb-b523-41ac7ee1dd5b

Parkrun 1280x720 - 50fps, GOP = 250

minih264:

https://github.com/user-attachments/assets/bcc408bd-6d5b-4725-b26e-307eee21ba56

x264 (encoded via ffmpeg with -c:v libx264 -preset superfast -crf 29.8 -refs 1 -coder vlc -bf 0 -g 250):

https://github.com/user-attachments/assets/a59c4027-4d6e-4b82-929b-4e344ca931d8

DSV2: (Note this video here had to be re-encoded to fit under 10MB. I re-encoded with x264 CRF 22.5 with the veryslow preset)

https://github.com/user-attachments/assets/66f91774-be8b-4c94-8cce-68e27f98c5c8