SlimeCodec

Commutator-norm driven video compression — block-level bit allocation via NormMAP

Japanese / 日本語

Commutator Norm as a Continuous Criterion for GOP Safety and Perceptual Quality Redistribution in Video Encoding
Hiroshi Sasaki — submitted to IEEE TCSVT, 2026

Empirical, May 2026

−74.3%
4K H.264 → 1/4 size
VMAF 83.25
visually identical
54 KB
WASM 4K AV1 decoder, 2-core CPU

Paper claim: −52 to −65% file size at VMAF 83–88, −1.62% VMAF BD-rate at 1080p. 100% scene-cut detection within ±2 frames.

4K side-by-side: original vs NormMAP H.264 1/4 size

Same 4K nature frame · left: original 110 MB · right: NormMAP H.264 28 MB (−74%, VMAF 83.25)

📢 Production trial available: javatel.co.jp/slimecodec/ (NormH.264, 30-day HW-locked) · /slimecodec-av1/ (NormAV1, research preview)
Paper

SlimeCodec_TCSVT_v8.pdf

Commutator Norm as a Continuous Criterion for GOP Safety and Perceptual Quality Redistribution in Video Encoding
Hiroshi Sasaki — TCSVT v8 (2026-04-18)

366 KB / IEEE TCSVT format / 14 pp

DOWNLOAD .pdf
Encoder binaries
Latest production-grade encoder (NormH.264 / NormAV1 v0.3.3) is distributed at the commercial trial site to keep this academic landing's WASM demos self-consistent with the paper's .normmap format snapshot.

→ javatel.co.jp/slimecodec/ (Windows GUI/CLI + Linux, 30-day HW-locked trial)
→ javatel.co.jp/slimecodec-av1/ (NormAV1 SVT-AV1 ROI map, research preview)

The paper-cited .normmap reproduction snapshot stays co-located with these WASM demos to keep them self-consistent.
Browser demos (WASM, no install)

WASM 2-core 4K AV1 Decoder

The paper's “2-core WASM 4K AV1 real-time decode” claim, playable in browser. Verified on Apple A15 (iPhone 13 ARM 2C), Intel i5-1145G7 (x86 mobile 2C) and AMD Ryzen 5 GHz desktop. Phase-synchronized parallel decode is observed end-to-end via dav1d frame threading.

javatel.co.jp · open in browser
Applications & roadmap
Beyond bit-allocation, NormMAP can be extended via integral geometry (IGP) into a decoder-side semantic stream — bounding boxes, masks and class labels emerge from the commutator-norm field with negligible overhead.

Current and near-term applications:
• Surveillance / VMS / NVR — on-device object detection + bbox tracking, archival under low bandwidth (LTE 4G fits 4K).
• Edge AI pipelines — decoder emits semantic stream, not raw frames; downstream AI input is orders of magnitude lighter.
• Smart-city / traffic — vehicle and pedestrian counts derived during decode, no second-pass inference.
• Drone / aerial — change detection and anomaly flags on the airborne side; only deltas uploaded.
• Live broadcast — player / ball / object tracking as a free overlay layer synchronised with display.
• Mobile / IoT — battery-friendly because the expensive part (semantic extraction) reuses commutator-norm already needed by the decode path.

The 2-core WASM 4K AV1 decoder doubles as the platform reference: any device that plays the demo can host the NormMAP × IGP pipeline.
Citation (BibTeX)
@article{sasaki2026slimecodec, author = {Sasaki, Hiroshi}, title = {Commutator Norm as a Continuous Criterion for GOP Safety and Perceptual Quality Redistribution in Video Encoding}, journal = {submitted to IEEE Trans. Circuits Syst. Video Technol.}, year = {2026}, note = {SlimeCodec NormMAP, Patent pending JP 2026-046609} }