rate-distortion

This repository contains scripts to calculate the famous Bjontegaard-Delta metric with different interpolation functions.

Matlab code to compute entropy, conditional entropy, mutual information, channel capacity, and rate distortion.

Official PyTorch implementation of AFP-GIC: controllable generative image compression with a single model for multiple bitrates.

Reproducible research toolkit for benchmarking AVIF, JPEG XL, WebP, and JPEG on web images with perceptual quality metrics, statistical analysis, visual comparisons, and interactive reports.

Neural Distributed Compressor Discovers Binning [JSAIT 2024]

Implementations of different ways to calculate Shannon's rate-distortion trade-off

Observation Theory — registration-first claim ledger, prediction registry, experiments, and the foundational paper (Book XIV of the Geometric Series).

A source-aware rate-distortion index of image and video tokenizers.

Researching efficient, reliable communication between LLM agents without assuming a new language wins.

Observation Theory — the front door: thesis, method, and the map of the ledgered repositories. The theory has one public name; every claim resolves to a ledger row.

Rigorous bounds on the gap between learned neural codec rate and the optimal rate-distortion function R(D).

Information, compression, coding, capacity, rate-distortion and neural estimation with reproducible experiments and interactive comparisons.

Honest accounting of generative (model-as-codec) compression: VAE video, EnCodec/DAC audio, MIDI — with the shared model counted as cost. 모델을 공유하면 전송량은 얼마나 줄고 무엇을 잃는가

The following project is an analytical analysis which compares deep learned image compression (CompressAI, Conv2D Autoencoder) against traditional codecs (JPEG, PNG) on CIFAR-10 on the basis of the benchmarks: rate-distortion and latency.

Three MQTT data-transmission systems using learned and classical compression — latent-space image codec, Relative Entropy Coding, and REC over a VAE latent — each with sender/receiver apps, full analysis, and documentation.

Honest RD evaluation for compression research: gate-checked Bjontegaard (BD-rate) metrics, matched-quality single-point comparison, publication-style RD plots

MATLAB image compression: AEV-DLT v3 per-coefficient fair-rate PCRD (CDF 9/7 wavelet) - beats EBCOT Tier-2 baseline on boat & peppers

Benchmark of the VAEs in Wan2.1, Wan2.2, LTX-Video, and HunyuanVideo against x264, x265, and SVT-AV1.

Code for signal recovery from quantized sparsified measurements with imperfect side information

Extends high-rate asymptotic theory of Nonlinear Transform Coding (NTC) to finite-rate regimes with numerically verified bounds.