Towards 1000-fold Electron Microscopy Image Compression for Connectomics via VQ-VAE with Transformer Prior

2025-10-31

AI Summary

To address storage, transmission, and analysis bottlenecks posed by petabyte-scale electron microscopy (EM) images in connectomics, this paper proposes an ROI-aware, variable-rate VQ-VAE compression framework. Methodologically: (1) a Transformer-based prior network predicts discrete codebook indices to enhance symbol modeling; (2) FiLM conditioning and feature concatenation are integrated into the decoder to improve texture reconstruction fidelity without altering the compression ratio; and (3) an ROI-driven selective decoding pipeline enables on-demand hierarchical decompression and localized high-resolution reconstruction. Experiments demonstrate semantic usability across variable compression ratios from 16× to 1024×, significantly reducing data overhead while simultaneously achieving high global compression efficiency and fine-grained structural recovery in critical regions.

Technology Category

• Data Mining & Knowledge Management: Data Compression
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Computer Vision: Medical and Biological Imaging

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Petascale electron microscopy (EM) datasets push storage, transfer, and downstream analysis toward their current limits. We present a vector-quantized variational autoencoder-based (VQ-VAE) compression framework for EM that spans 16x to 1024x and enables pay-as-you-decode usage: top-only decoding for extreme compression, with an optional Transformer prior that predicts bottom tokens (without changing the compression ratio) to restore texture via feature-wise linear modulation (FiLM) and concatenation; we further introduce an ROI-driven workflow that performs selective high-resolution reconstruction from 1024x-compressed latents only where needed.

Problem

• Compressing petascale electron microscopy datasets efficiently
• Enabling scalable decoding with optional texture restoration
• Providing selective high-resolution reconstruction from compressed data

Innovation

• VQ-VAE framework enables scalable EM image compression
• Transformer prior restores texture without altering compression ratio
• ROI-driven workflow reconstructs high-resolution selectively from compressed latents