I just generated 1,000,000 transaction graph visualizations from real Ethereum/Arbitrum/Polygon data — now training a Vision-Language model to detect DeFi attacks

What is the 1M Transaction Graph Visualization Dataset?

A developer just open-sourced a **1M+ transaction graph visualization dataset** from real Ethereum/Arbitrum/Polygon data, training vision-language models to spot DeFi attacks through visual pattern recognition.

- 1M visual transaction graph images from 1.87M on-chain transactions

How Vision-Language Models Detect DeFi Attacks

- Attack topology labels: `DRAIN_STAR`, `MIXING_CHAIN`, `NORMAL`

- Generated in 1h15min using 20-core AMD MI300X parallel processing

Real-World Applications and Security Implications

- Fine-tuned **Qwen2-VL-7B** via LoRA on AMD ROCm

- **ERC-8259** proposal for AI Agent Identity & Threat Registry

- **Integration**: Build on ERC-8259 standard for AI agent security

- **Research**: Extend visual analysis to other attack vectors

Imina-Na V2 vision brain deployment, ERC-8259 standardization process, and expanding graph generation to more chains. The Sigui DePIN security oracle aims to provide real-time visual threat detection for autonomous agents.

Open dataset means anyone can build competing models or extend this approach to new attack types.