feat(paddleocr-vl): add PaddleOCR-VL full pipeline Docker image and API server, plus integration tests and docker helpers
This commit is contained in:
297
test/helpers/docker.ts
Normal file
297
test/helpers/docker.ts
Normal file
@@ -0,0 +1,297 @@
|
||||
import { execSync } from 'child_process';
|
||||
|
||||
// Project container names (only manage these)
|
||||
const PROJECT_CONTAINERS = [
|
||||
'paddleocr-vl-test',
|
||||
'paddleocr-vl-gpu-test',
|
||||
'paddleocr-vl-cpu-test',
|
||||
'paddleocr-vl-full-test',
|
||||
'minicpm-test',
|
||||
];
|
||||
|
||||
// Image configurations
|
||||
export interface IImageConfig {
|
||||
name: string;
|
||||
dockerfile: string;
|
||||
buildContext: string;
|
||||
containerName: string;
|
||||
ports: string[];
|
||||
volumes?: string[];
|
||||
gpus?: boolean;
|
||||
healthEndpoint?: string;
|
||||
healthTimeout?: number;
|
||||
}
|
||||
|
||||
export const IMAGES = {
|
||||
paddleocrVlGpu: {
|
||||
name: 'paddleocr-vl-gpu',
|
||||
dockerfile: 'Dockerfile_paddleocr_vl_gpu',
|
||||
buildContext: '.',
|
||||
containerName: 'paddleocr-vl-test',
|
||||
ports: ['8000:8000'],
|
||||
volumes: ['ht-huggingface-cache:/root/.cache/huggingface'],
|
||||
gpus: true,
|
||||
healthEndpoint: 'http://localhost:8000/health',
|
||||
healthTimeout: 300000, // 5 minutes for model loading
|
||||
} as IImageConfig,
|
||||
|
||||
paddleocrVlCpu: {
|
||||
name: 'paddleocr-vl-cpu',
|
||||
dockerfile: 'Dockerfile_paddleocr_vl_cpu',
|
||||
buildContext: '.',
|
||||
containerName: 'paddleocr-vl-test',
|
||||
ports: ['8000:8000'],
|
||||
volumes: ['ht-huggingface-cache:/root/.cache/huggingface'],
|
||||
gpus: false,
|
||||
healthEndpoint: 'http://localhost:8000/health',
|
||||
healthTimeout: 300000,
|
||||
} as IImageConfig,
|
||||
|
||||
minicpm: {
|
||||
name: 'minicpm45v',
|
||||
dockerfile: 'Dockerfile_minicpm45v',
|
||||
buildContext: '.',
|
||||
containerName: 'minicpm-test',
|
||||
ports: ['11434:11434'],
|
||||
volumes: ['ht-ollama-models:/root/.ollama'],
|
||||
gpus: true,
|
||||
healthEndpoint: 'http://localhost:11434/api/tags',
|
||||
healthTimeout: 120000,
|
||||
} as IImageConfig,
|
||||
|
||||
// Full PaddleOCR-VL pipeline with PP-DocLayoutV2 + structured JSON output
|
||||
paddleocrVlFull: {
|
||||
name: 'paddleocr-vl-full',
|
||||
dockerfile: 'Dockerfile_paddleocr_vl_full',
|
||||
buildContext: '.',
|
||||
containerName: 'paddleocr-vl-full-test',
|
||||
ports: ['8000:8000'],
|
||||
volumes: [
|
||||
'ht-huggingface-cache:/root/.cache/huggingface',
|
||||
'ht-paddleocr-cache:/root/.paddleocr',
|
||||
],
|
||||
gpus: true,
|
||||
healthEndpoint: 'http://localhost:8000/health',
|
||||
healthTimeout: 600000, // 10 minutes for model loading (vLLM + PP-DocLayoutV2)
|
||||
} as IImageConfig,
|
||||
};
|
||||
|
||||
/**
|
||||
* Execute a shell command and return output
|
||||
*/
|
||||
function exec(command: string, silent = false): string {
|
||||
try {
|
||||
return execSync(command, {
|
||||
encoding: 'utf-8',
|
||||
stdio: silent ? 'pipe' : 'inherit',
|
||||
});
|
||||
} catch (err: unknown) {
|
||||
if (silent) return '';
|
||||
throw err;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if a Docker image exists locally
|
||||
*/
|
||||
export function imageExists(imageName: string): boolean {
|
||||
const result = exec(`docker images -q ${imageName}`, true);
|
||||
return result.trim().length > 0;
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if a container is running
|
||||
*/
|
||||
export function isContainerRunning(containerName: string): boolean {
|
||||
const result = exec(`docker ps --filter "name=^${containerName}$" --format "{{.Names}}"`, true);
|
||||
return result.trim() === containerName;
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if a container exists (running or stopped)
|
||||
*/
|
||||
export function containerExists(containerName: string): boolean {
|
||||
const result = exec(`docker ps -a --filter "name=^${containerName}$" --format "{{.Names}}"`, true);
|
||||
return result.trim() === containerName;
|
||||
}
|
||||
|
||||
/**
|
||||
* Stop and remove a container
|
||||
*/
|
||||
export function removeContainer(containerName: string): void {
|
||||
if (containerExists(containerName)) {
|
||||
console.log(`[Docker] Removing container: ${containerName}`);
|
||||
exec(`docker rm -f ${containerName}`, true);
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Stop all project containers that conflict with the required one
|
||||
*/
|
||||
export function stopConflictingContainers(requiredContainer: string, requiredPort: string): void {
|
||||
// Stop project containers using the same port
|
||||
for (const container of PROJECT_CONTAINERS) {
|
||||
if (container === requiredContainer) continue;
|
||||
|
||||
if (isContainerRunning(container)) {
|
||||
// Check if this container uses the same port
|
||||
const ports = exec(`docker port ${container} 2>/dev/null || true`, true);
|
||||
if (ports.includes(requiredPort.split(':')[0])) {
|
||||
console.log(`[Docker] Stopping conflicting container: ${container}`);
|
||||
exec(`docker stop ${container}`, true);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Build a Docker image
|
||||
*/
|
||||
export function buildImage(config: IImageConfig): void {
|
||||
console.log(`[Docker] Building image: ${config.name}`);
|
||||
const cmd = `docker build --load -f ${config.dockerfile} -t ${config.name} ${config.buildContext}`;
|
||||
exec(cmd);
|
||||
}
|
||||
|
||||
/**
|
||||
* Start a container from an image
|
||||
*/
|
||||
export function startContainer(config: IImageConfig): void {
|
||||
// Remove existing container if it exists
|
||||
removeContainer(config.containerName);
|
||||
|
||||
console.log(`[Docker] Starting container: ${config.containerName}`);
|
||||
|
||||
const portArgs = config.ports.map((p) => `-p ${p}`).join(' ');
|
||||
const volumeArgs = config.volumes?.map((v) => `-v ${v}`).join(' ') || '';
|
||||
const gpuArgs = config.gpus ? '--gpus all' : '';
|
||||
|
||||
const cmd = `docker run -d --name ${config.containerName} ${gpuArgs} ${portArgs} ${volumeArgs} ${config.name}`;
|
||||
exec(cmd);
|
||||
}
|
||||
|
||||
/**
|
||||
* Wait for a container to become healthy
|
||||
*/
|
||||
export async function waitForHealth(
|
||||
endpoint: string,
|
||||
timeoutMs: number = 120000,
|
||||
intervalMs: number = 5000
|
||||
): Promise<boolean> {
|
||||
const startTime = Date.now();
|
||||
console.log(`[Docker] Waiting for health: ${endpoint}`);
|
||||
|
||||
while (Date.now() - startTime < timeoutMs) {
|
||||
try {
|
||||
const response = await fetch(endpoint, {
|
||||
method: 'GET',
|
||||
signal: AbortSignal.timeout(5000),
|
||||
});
|
||||
if (response.ok) {
|
||||
console.log(`[Docker] Service healthy!`);
|
||||
return true;
|
||||
}
|
||||
} catch {
|
||||
// Service not ready yet
|
||||
}
|
||||
|
||||
const elapsed = Math.round((Date.now() - startTime) / 1000);
|
||||
console.log(`[Docker] Waiting... (${elapsed}s)`);
|
||||
await new Promise((resolve) => setTimeout(resolve, intervalMs));
|
||||
}
|
||||
|
||||
console.log(`[Docker] Health check timeout after ${timeoutMs / 1000}s`);
|
||||
return false;
|
||||
}
|
||||
|
||||
/**
|
||||
* Ensure a service is running and healthy
|
||||
* - Builds image if missing
|
||||
* - Stops conflicting project containers
|
||||
* - Starts container if not running
|
||||
* - Waits for health check
|
||||
*/
|
||||
export async function ensureService(config: IImageConfig): Promise<boolean> {
|
||||
console.log(`\n[Docker] Ensuring service: ${config.name}`);
|
||||
|
||||
// Build image if it doesn't exist
|
||||
if (!imageExists(config.name)) {
|
||||
console.log(`[Docker] Image not found, building...`);
|
||||
buildImage(config);
|
||||
}
|
||||
|
||||
// Stop conflicting containers on the same port
|
||||
const mainPort = config.ports[0];
|
||||
stopConflictingContainers(config.containerName, mainPort);
|
||||
|
||||
// Start container if not running
|
||||
if (!isContainerRunning(config.containerName)) {
|
||||
startContainer(config);
|
||||
} else {
|
||||
console.log(`[Docker] Container already running: ${config.containerName}`);
|
||||
}
|
||||
|
||||
// Wait for health
|
||||
if (config.healthEndpoint) {
|
||||
return waitForHealth(config.healthEndpoint, config.healthTimeout);
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
/**
|
||||
* Ensure PaddleOCR-VL GPU service is running
|
||||
*/
|
||||
export async function ensurePaddleOcrVlGpu(): Promise<boolean> {
|
||||
return ensureService(IMAGES.paddleocrVlGpu);
|
||||
}
|
||||
|
||||
/**
|
||||
* Ensure PaddleOCR-VL CPU service is running
|
||||
*/
|
||||
export async function ensurePaddleOcrVlCpu(): Promise<boolean> {
|
||||
return ensureService(IMAGES.paddleocrVlCpu);
|
||||
}
|
||||
|
||||
/**
|
||||
* Ensure MiniCPM service is running
|
||||
*/
|
||||
export async function ensureMiniCpm(): Promise<boolean> {
|
||||
return ensureService(IMAGES.minicpm);
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if GPU is available
|
||||
*/
|
||||
export function isGpuAvailable(): boolean {
|
||||
try {
|
||||
const result = exec('nvidia-smi --query-gpu=name --format=csv,noheader 2>/dev/null', true);
|
||||
return result.trim().length > 0;
|
||||
} catch {
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Ensure PaddleOCR-VL service (auto-detect GPU/CPU)
|
||||
*/
|
||||
export async function ensurePaddleOcrVl(): Promise<boolean> {
|
||||
if (isGpuAvailable()) {
|
||||
console.log('[Docker] GPU detected, using GPU image');
|
||||
return ensurePaddleOcrVlGpu();
|
||||
} else {
|
||||
console.log('[Docker] No GPU detected, using CPU image');
|
||||
return ensurePaddleOcrVlCpu();
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Ensure PaddleOCR-VL Full Pipeline service (PP-DocLayoutV2 + structured output)
|
||||
* This is the recommended service for production use - outputs structured JSON/Markdown
|
||||
*/
|
||||
export async function ensurePaddleOcrVlFull(): Promise<boolean> {
|
||||
if (!isGpuAvailable()) {
|
||||
console.log('[Docker] WARNING: Full pipeline requires GPU, but none detected');
|
||||
}
|
||||
return ensureService(IMAGES.paddleocrVlFull);
|
||||
}
|
||||
@@ -1,15 +1,23 @@
|
||||
/**
|
||||
* Bank statement extraction test using MiniCPM-V (visual) + PaddleOCR-VL (table recognition)
|
||||
*
|
||||
* This is the combined/dual-VLM approach that uses both models for consensus:
|
||||
* - MiniCPM-V for visual extraction
|
||||
* - PaddleOCR-VL for table recognition
|
||||
*/
|
||||
import { tap, expect } from '@git.zone/tstest/tapbundle';
|
||||
import * as fs from 'fs';
|
||||
import * as path from 'path';
|
||||
import { execSync } from 'child_process';
|
||||
import * as os from 'os';
|
||||
import { ensurePaddleOcrVl, ensureMiniCpm } from './helpers/docker.js';
|
||||
|
||||
// Service URLs
|
||||
const OLLAMA_URL = 'http://localhost:11434';
|
||||
const PADDLEOCR_VL_URL = 'http://localhost:8000';
|
||||
|
||||
// Models
|
||||
const MINICPM_MODEL = 'openbmb/minicpm-v4.5:q8_0';
|
||||
const MINICPM_MODEL = 'minicpm-v:latest';
|
||||
const PADDLEOCR_VL_MODEL = 'paddleocr-vl';
|
||||
|
||||
// Prompt for MiniCPM-V visual extraction
|
||||
@@ -477,11 +485,18 @@ function findTestCases(): Array<{ name: string; pdfPath: string; jsonPath: strin
|
||||
|
||||
// Tests
|
||||
|
||||
tap.test('should connect to Ollama API', async () => {
|
||||
const response = await fetch(`${OLLAMA_URL}/api/tags`);
|
||||
expect(response.ok).toBeTrue();
|
||||
const data = await response.json();
|
||||
expect(data.models).toBeArray();
|
||||
tap.test('setup: ensure Docker containers are running', async () => {
|
||||
console.log('\n[Setup] Checking Docker containers...\n');
|
||||
|
||||
// Ensure PaddleOCR-VL is running (auto-detects GPU/CPU)
|
||||
const paddleOk = await ensurePaddleOcrVl();
|
||||
expect(paddleOk).toBeTrue();
|
||||
|
||||
// Ensure MiniCPM is running
|
||||
const minicpmOk = await ensureMiniCpm();
|
||||
expect(minicpmOk).toBeTrue();
|
||||
|
||||
console.log('\n[Setup] All containers ready!\n');
|
||||
});
|
||||
|
||||
tap.test('should have MiniCPM-V 4.5 model loaded', async () => {
|
||||
@@ -494,8 +509,7 @@ tap.test('should have MiniCPM-V 4.5 model loaded', async () => {
|
||||
tap.test('should check PaddleOCR-VL availability', async () => {
|
||||
const available = await isPaddleOCRVLAvailable();
|
||||
console.log(`PaddleOCR-VL available: ${available}`);
|
||||
// This test passes regardless - PaddleOCR-VL is optional
|
||||
expect(true).toBeTrue();
|
||||
expect(available).toBeTrue();
|
||||
});
|
||||
|
||||
// Dynamic test for each PDF/JSON pair
|
||||
|
||||
334
test/test.bankstatements.minicpm.ts
Normal file
334
test/test.bankstatements.minicpm.ts
Normal file
@@ -0,0 +1,334 @@
|
||||
/**
|
||||
* Bank statement extraction test using MiniCPM-V only (visual extraction)
|
||||
*
|
||||
* This tests MiniCPM-V's ability to extract bank transactions directly from images
|
||||
* without any OCR augmentation.
|
||||
*/
|
||||
import { tap, expect } from '@git.zone/tstest/tapbundle';
|
||||
import * as fs from 'fs';
|
||||
import * as path from 'path';
|
||||
import { execSync } from 'child_process';
|
||||
import * as os from 'os';
|
||||
import { ensureMiniCpm } from './helpers/docker.js';
|
||||
|
||||
// Service URL
|
||||
const OLLAMA_URL = 'http://localhost:11434';
|
||||
|
||||
// Model
|
||||
const MINICPM_MODEL = 'minicpm-v:latest';
|
||||
|
||||
// Prompt for MiniCPM-V visual extraction
|
||||
const MINICPM_EXTRACT_PROMPT = `/nothink
|
||||
You are a bank statement parser. Extract EVERY transaction from the table.
|
||||
|
||||
Read the Amount column carefully:
|
||||
- "- 21,47 €" means DEBIT, output as: -21.47
|
||||
- "+ 1.000,00 €" means CREDIT, output as: 1000.00
|
||||
- European format: comma = decimal point
|
||||
|
||||
For each row output: {"date":"YYYY-MM-DD","counterparty":"NAME","amount":-21.47}
|
||||
|
||||
Do not skip any rows. Return ONLY the JSON array, no explanation.`;
|
||||
|
||||
interface ITransaction {
|
||||
date: string;
|
||||
counterparty: string;
|
||||
amount: number;
|
||||
}
|
||||
|
||||
/**
|
||||
* Convert PDF to PNG images using ImageMagick
|
||||
*/
|
||||
function convertPdfToImages(pdfPath: string): string[] {
|
||||
const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'pdf-convert-'));
|
||||
const outputPattern = path.join(tempDir, 'page-%d.png');
|
||||
|
||||
try {
|
||||
execSync(
|
||||
`convert -density 300 -quality 100 "${pdfPath}" -background white -alpha remove "${outputPattern}"`,
|
||||
{ stdio: 'pipe' }
|
||||
);
|
||||
|
||||
const files = fs.readdirSync(tempDir).filter((f: string) => f.endsWith('.png')).sort();
|
||||
const images: string[] = [];
|
||||
|
||||
for (const file of files) {
|
||||
const imagePath = path.join(tempDir, file);
|
||||
const imageData = fs.readFileSync(imagePath);
|
||||
images.push(imageData.toString('base64'));
|
||||
}
|
||||
|
||||
return images;
|
||||
} finally {
|
||||
fs.rmSync(tempDir, { recursive: true, force: true });
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Extract using MiniCPM-V via Ollama
|
||||
*/
|
||||
async function extractWithMiniCPM(images: string[], passLabel: string): Promise<ITransaction[]> {
|
||||
const payload = {
|
||||
model: MINICPM_MODEL,
|
||||
prompt: MINICPM_EXTRACT_PROMPT,
|
||||
images,
|
||||
stream: true,
|
||||
options: {
|
||||
num_predict: 16384,
|
||||
temperature: 0.1,
|
||||
},
|
||||
};
|
||||
|
||||
const response = await fetch(`${OLLAMA_URL}/api/generate`, {
|
||||
method: 'POST',
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
body: JSON.stringify(payload),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
throw new Error(`Ollama API error: ${response.status}`);
|
||||
}
|
||||
|
||||
const reader = response.body?.getReader();
|
||||
if (!reader) {
|
||||
throw new Error('No response body');
|
||||
}
|
||||
|
||||
const decoder = new TextDecoder();
|
||||
let fullText = '';
|
||||
let lineBuffer = '';
|
||||
|
||||
console.log(`[${passLabel}] Extracting with MiniCPM-V...`);
|
||||
|
||||
while (true) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) break;
|
||||
|
||||
const chunk = decoder.decode(value, { stream: true });
|
||||
const lines = chunk.split('\n').filter((l) => l.trim());
|
||||
|
||||
for (const line of lines) {
|
||||
try {
|
||||
const json = JSON.parse(line);
|
||||
if (json.response) {
|
||||
fullText += json.response;
|
||||
lineBuffer += json.response;
|
||||
|
||||
if (lineBuffer.includes('\n')) {
|
||||
const parts = lineBuffer.split('\n');
|
||||
for (let i = 0; i < parts.length - 1; i++) {
|
||||
console.log(parts[i]);
|
||||
}
|
||||
lineBuffer = parts[parts.length - 1];
|
||||
}
|
||||
}
|
||||
} catch {
|
||||
// Skip invalid JSON lines
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (lineBuffer) {
|
||||
console.log(lineBuffer);
|
||||
}
|
||||
console.log('');
|
||||
|
||||
const startIdx = fullText.indexOf('[');
|
||||
const endIdx = fullText.lastIndexOf(']') + 1;
|
||||
|
||||
if (startIdx < 0 || endIdx <= startIdx) {
|
||||
throw new Error('No JSON array found in response');
|
||||
}
|
||||
|
||||
return JSON.parse(fullText.substring(startIdx, endIdx));
|
||||
}
|
||||
|
||||
/**
|
||||
* Create a hash of transactions for comparison
|
||||
*/
|
||||
function hashTransactions(transactions: ITransaction[]): string {
|
||||
return transactions
|
||||
.map((t) => `${t.date}|${t.amount.toFixed(2)}`)
|
||||
.sort()
|
||||
.join(';');
|
||||
}
|
||||
|
||||
/**
|
||||
* Extract with consensus voting using MiniCPM-V only
|
||||
*/
|
||||
async function extractWithConsensus(
|
||||
images: string[],
|
||||
maxPasses: number = 5
|
||||
): Promise<ITransaction[]> {
|
||||
const results: Array<{ transactions: ITransaction[]; hash: string }> = [];
|
||||
const hashCounts: Map<string, number> = new Map();
|
||||
|
||||
const addResult = (transactions: ITransaction[], passLabel: string): number => {
|
||||
const hash = hashTransactions(transactions);
|
||||
results.push({ transactions, hash });
|
||||
hashCounts.set(hash, (hashCounts.get(hash) || 0) + 1);
|
||||
console.log(
|
||||
`[${passLabel}] Got ${transactions.length} transactions (hash: ${hash.substring(0, 20)}...)`
|
||||
);
|
||||
return hashCounts.get(hash)!;
|
||||
};
|
||||
|
||||
console.log('[Setup] Using MiniCPM-V only');
|
||||
|
||||
for (let pass = 1; pass <= maxPasses; pass++) {
|
||||
try {
|
||||
const transactions = await extractWithMiniCPM(images, `Pass ${pass} MiniCPM-V`);
|
||||
const count = addResult(transactions, `Pass ${pass} MiniCPM-V`);
|
||||
|
||||
if (count >= 2) {
|
||||
console.log(`[Consensus] Reached after ${pass} passes`);
|
||||
return transactions;
|
||||
}
|
||||
|
||||
console.log(`[Pass ${pass}] No consensus yet, trying again...`);
|
||||
} catch (err) {
|
||||
console.log(`[Pass ${pass}] Error: ${err}`);
|
||||
}
|
||||
}
|
||||
|
||||
// No consensus reached - return the most common result
|
||||
let bestHash = '';
|
||||
let bestCount = 0;
|
||||
for (const [hash, count] of hashCounts) {
|
||||
if (count > bestCount) {
|
||||
bestCount = count;
|
||||
bestHash = hash;
|
||||
}
|
||||
}
|
||||
|
||||
if (!bestHash) {
|
||||
throw new Error('No valid results obtained');
|
||||
}
|
||||
|
||||
const best = results.find((r) => r.hash === bestHash)!;
|
||||
console.log(`[No consensus] Using most common result (${bestCount}/${maxPasses} passes)`);
|
||||
return best.transactions;
|
||||
}
|
||||
|
||||
/**
|
||||
* Compare extracted transactions against expected
|
||||
*/
|
||||
function compareTransactions(
|
||||
extracted: ITransaction[],
|
||||
expected: ITransaction[]
|
||||
): { matches: number; total: number; errors: string[] } {
|
||||
const errors: string[] = [];
|
||||
let matches = 0;
|
||||
|
||||
for (let i = 0; i < expected.length; i++) {
|
||||
const exp = expected[i];
|
||||
const ext = extracted[i];
|
||||
|
||||
if (!ext) {
|
||||
errors.push(`Missing transaction ${i}: ${exp.date} ${exp.counterparty}`);
|
||||
continue;
|
||||
}
|
||||
|
||||
const dateMatch = ext.date === exp.date;
|
||||
const amountMatch = Math.abs(ext.amount - exp.amount) < 0.01;
|
||||
|
||||
if (dateMatch && amountMatch) {
|
||||
matches++;
|
||||
} else {
|
||||
errors.push(
|
||||
`Mismatch at ${i}: expected ${exp.date}/${exp.amount}, got ${ext.date}/${ext.amount}`
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
if (extracted.length > expected.length) {
|
||||
errors.push(`Extra transactions: ${extracted.length - expected.length}`);
|
||||
}
|
||||
|
||||
return { matches, total: expected.length, errors };
|
||||
}
|
||||
|
||||
/**
|
||||
* Find all test cases (PDF + JSON pairs) in .nogit/
|
||||
*/
|
||||
function findTestCases(): Array<{ name: string; pdfPath: string; jsonPath: string }> {
|
||||
const testDir = path.join(process.cwd(), '.nogit');
|
||||
if (!fs.existsSync(testDir)) {
|
||||
return [];
|
||||
}
|
||||
|
||||
const files = fs.readdirSync(testDir);
|
||||
const pdfFiles = files.filter((f: string) => f.endsWith('.pdf'));
|
||||
const testCases: Array<{ name: string; pdfPath: string; jsonPath: string }> = [];
|
||||
|
||||
for (const pdf of pdfFiles) {
|
||||
const baseName = pdf.replace('.pdf', '');
|
||||
const jsonFile = `${baseName}.json`;
|
||||
if (files.includes(jsonFile)) {
|
||||
testCases.push({
|
||||
name: baseName,
|
||||
pdfPath: path.join(testDir, pdf),
|
||||
jsonPath: path.join(testDir, jsonFile),
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
return testCases;
|
||||
}
|
||||
|
||||
// Tests
|
||||
|
||||
tap.test('setup: ensure Docker containers are running', async () => {
|
||||
console.log('\n[Setup] Checking Docker containers...\n');
|
||||
|
||||
// Ensure MiniCPM is running
|
||||
const minicpmOk = await ensureMiniCpm();
|
||||
expect(minicpmOk).toBeTrue();
|
||||
|
||||
console.log('\n[Setup] All containers ready!\n');
|
||||
});
|
||||
|
||||
tap.test('should have MiniCPM-V 4.5 model loaded', async () => {
|
||||
const response = await fetch(`${OLLAMA_URL}/api/tags`);
|
||||
const data = await response.json();
|
||||
const modelNames = data.models.map((m: { name: string }) => m.name);
|
||||
expect(modelNames.some((name: string) => name.includes('minicpm-v4.5'))).toBeTrue();
|
||||
});
|
||||
|
||||
// Dynamic test for each PDF/JSON pair
|
||||
const testCases = findTestCases();
|
||||
console.log(`\nFound ${testCases.length} bank statement test cases (MiniCPM-V only)\n`);
|
||||
|
||||
for (const testCase of testCases) {
|
||||
tap.test(`should extract transactions from ${testCase.name}`, async () => {
|
||||
// Load expected transactions
|
||||
const expected: ITransaction[] = JSON.parse(fs.readFileSync(testCase.jsonPath, 'utf-8'));
|
||||
console.log(`\n=== ${testCase.name} ===`);
|
||||
console.log(`Expected: ${expected.length} transactions`);
|
||||
|
||||
// Convert PDF to images
|
||||
console.log('Converting PDF to images...');
|
||||
const images = convertPdfToImages(testCase.pdfPath);
|
||||
console.log(`Converted: ${images.length} pages\n`);
|
||||
|
||||
// Extract with consensus (MiniCPM-V only)
|
||||
const extracted = await extractWithConsensus(images);
|
||||
console.log(`\nFinal: ${extracted.length} transactions`);
|
||||
|
||||
// Compare results
|
||||
const result = compareTransactions(extracted, expected);
|
||||
console.log(`Accuracy: ${result.matches}/${result.total}`);
|
||||
|
||||
if (result.errors.length > 0) {
|
||||
console.log('Errors:');
|
||||
result.errors.forEach((e) => console.log(` - ${e}`));
|
||||
}
|
||||
|
||||
// Assert high accuracy
|
||||
const accuracy = result.matches / result.total;
|
||||
expect(accuracy).toBeGreaterThan(0.95);
|
||||
expect(extracted.length).toEqual(expected.length);
|
||||
});
|
||||
}
|
||||
|
||||
export default tap.start();
|
||||
346
test/test.bankstatements.paddleocr-vl.ts
Normal file
346
test/test.bankstatements.paddleocr-vl.ts
Normal file
@@ -0,0 +1,346 @@
|
||||
/**
|
||||
* Bank statement extraction test using PaddleOCR-VL Full Pipeline
|
||||
*
|
||||
* This tests the complete PaddleOCR-VL pipeline for bank statements:
|
||||
* 1. PP-DocLayoutV2 for layout detection
|
||||
* 2. PaddleOCR-VL for recognition (tables with proper structure)
|
||||
* 3. Structured Markdown output with tables
|
||||
* 4. MiniCPM extracts transactions from structured tables
|
||||
*
|
||||
* The structured Markdown has properly formatted tables,
|
||||
* making it much easier for MiniCPM to extract transaction data.
|
||||
*/
|
||||
import { tap, expect } from '@git.zone/tstest/tapbundle';
|
||||
import * as fs from 'fs';
|
||||
import * as path from 'path';
|
||||
import { execSync } from 'child_process';
|
||||
import * as os from 'os';
|
||||
import { ensurePaddleOcrVlFull, ensureMiniCpm } from './helpers/docker.js';
|
||||
|
||||
const PADDLEOCR_VL_URL = 'http://localhost:8000';
|
||||
const OLLAMA_URL = 'http://localhost:11434';
|
||||
const MINICPM_MODEL = 'minicpm-v:latest';
|
||||
|
||||
interface ITransaction {
|
||||
date: string;
|
||||
counterparty: string;
|
||||
amount: number;
|
||||
}
|
||||
|
||||
/**
|
||||
* Convert PDF to PNG images using ImageMagick
|
||||
*/
|
||||
function convertPdfToImages(pdfPath: string): string[] {
|
||||
const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'pdf-convert-'));
|
||||
const outputPattern = path.join(tempDir, 'page-%d.png');
|
||||
|
||||
try {
|
||||
execSync(
|
||||
`convert -density 300 -quality 100 "${pdfPath}" -background white -alpha remove "${outputPattern}"`,
|
||||
{ stdio: 'pipe' }
|
||||
);
|
||||
|
||||
const files = fs.readdirSync(tempDir).filter((f: string) => f.endsWith('.png')).sort();
|
||||
const images: string[] = [];
|
||||
|
||||
for (const file of files) {
|
||||
const imagePath = path.join(tempDir, file);
|
||||
const imageData = fs.readFileSync(imagePath);
|
||||
images.push(imageData.toString('base64'));
|
||||
}
|
||||
|
||||
return images;
|
||||
} finally {
|
||||
fs.rmSync(tempDir, { recursive: true, force: true });
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Parse document using PaddleOCR-VL Full Pipeline (returns structured Markdown)
|
||||
*/
|
||||
async function parseDocument(imageBase64: string): Promise<string> {
|
||||
const response = await fetch(`${PADDLEOCR_VL_URL}/parse`, {
|
||||
method: 'POST',
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
body: JSON.stringify({
|
||||
image: imageBase64,
|
||||
output_format: 'markdown',
|
||||
}),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
const text = await response.text();
|
||||
throw new Error(`PaddleOCR-VL API error: ${response.status} - ${text}`);
|
||||
}
|
||||
|
||||
const data = await response.json();
|
||||
|
||||
if (!data.success) {
|
||||
throw new Error(`PaddleOCR-VL error: ${data.error}`);
|
||||
}
|
||||
|
||||
return data.result?.markdown || '';
|
||||
}
|
||||
|
||||
/**
|
||||
* Extract transactions from structured Markdown using MiniCPM
|
||||
*/
|
||||
async function extractTransactionsFromMarkdown(markdown: string): Promise<ITransaction[]> {
|
||||
console.log(` [Extract] Processing ${markdown.length} chars of Markdown`);
|
||||
|
||||
const prompt = `/nothink
|
||||
Convert this bank statement to a JSON array of transactions.
|
||||
|
||||
Read the Amount values carefully:
|
||||
- "- 21,47 €" means DEBIT, output as: -21.47
|
||||
- "+ 1.000,00 €" means CREDIT, output as: 1000.00
|
||||
- European format: comma = decimal point, dot = thousands
|
||||
|
||||
For each transaction output: {"date":"YYYY-MM-DD","counterparty":"NAME","amount":-21.47}
|
||||
|
||||
Return ONLY the JSON array, no explanation.
|
||||
|
||||
Document:
|
||||
${markdown}`;
|
||||
|
||||
const payload = {
|
||||
model: MINICPM_MODEL,
|
||||
prompt,
|
||||
stream: true,
|
||||
options: {
|
||||
num_predict: 16384,
|
||||
temperature: 0.1,
|
||||
},
|
||||
};
|
||||
|
||||
const response = await fetch(`${OLLAMA_URL}/api/generate`, {
|
||||
method: 'POST',
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
body: JSON.stringify(payload),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
throw new Error(`Ollama API error: ${response.status}`);
|
||||
}
|
||||
|
||||
const reader = response.body?.getReader();
|
||||
if (!reader) {
|
||||
throw new Error('No response body');
|
||||
}
|
||||
|
||||
const decoder = new TextDecoder();
|
||||
let fullText = '';
|
||||
|
||||
while (true) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) break;
|
||||
|
||||
const chunk = decoder.decode(value, { stream: true });
|
||||
const lines = chunk.split('\n').filter((l) => l.trim());
|
||||
|
||||
for (const line of lines) {
|
||||
try {
|
||||
const json = JSON.parse(line);
|
||||
if (json.response) {
|
||||
fullText += json.response;
|
||||
}
|
||||
} catch {
|
||||
// Skip invalid JSON lines
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Extract JSON array from response
|
||||
const startIdx = fullText.indexOf('[');
|
||||
const endIdx = fullText.lastIndexOf(']') + 1;
|
||||
|
||||
if (startIdx < 0 || endIdx <= startIdx) {
|
||||
throw new Error(`No JSON array found in response: ${fullText.substring(0, 200)}`);
|
||||
}
|
||||
|
||||
const jsonStr = fullText.substring(startIdx, endIdx);
|
||||
return JSON.parse(jsonStr);
|
||||
}
|
||||
|
||||
/**
|
||||
* Extract transactions from all pages of a bank statement
|
||||
*/
|
||||
async function extractAllTransactions(images: string[]): Promise<ITransaction[]> {
|
||||
const allTransactions: ITransaction[] = [];
|
||||
|
||||
for (let i = 0; i < images.length; i++) {
|
||||
console.log(` Processing page ${i + 1}/${images.length}...`);
|
||||
|
||||
// Parse with full pipeline
|
||||
const markdown = await parseDocument(images[i]);
|
||||
console.log(` [Parse] Got ${markdown.split('\n').length} lines of Markdown`);
|
||||
|
||||
// Extract transactions
|
||||
try {
|
||||
const transactions = await extractTransactionsFromMarkdown(markdown);
|
||||
console.log(` [Extracted] ${transactions.length} transactions`);
|
||||
allTransactions.push(...transactions);
|
||||
} catch (err) {
|
||||
console.log(` [Error] ${err}`);
|
||||
}
|
||||
}
|
||||
|
||||
return allTransactions;
|
||||
}
|
||||
|
||||
/**
|
||||
* Compare transactions - find matching transaction in expected list
|
||||
*/
|
||||
function findMatchingTransaction(
|
||||
tx: ITransaction,
|
||||
expectedList: ITransaction[]
|
||||
): ITransaction | undefined {
|
||||
return expectedList.find((exp) => {
|
||||
const dateMatch = tx.date === exp.date;
|
||||
const amountMatch = Math.abs(tx.amount - exp.amount) < 0.02;
|
||||
const counterpartyMatch =
|
||||
tx.counterparty?.toLowerCase().includes(exp.counterparty?.toLowerCase().slice(0, 10)) ||
|
||||
exp.counterparty?.toLowerCase().includes(tx.counterparty?.toLowerCase().slice(0, 10));
|
||||
return dateMatch && amountMatch && counterpartyMatch;
|
||||
});
|
||||
}
|
||||
|
||||
/**
|
||||
* Calculate extraction accuracy
|
||||
*/
|
||||
function calculateAccuracy(
|
||||
extracted: ITransaction[],
|
||||
expected: ITransaction[]
|
||||
): { matched: number; total: number; accuracy: number } {
|
||||
let matched = 0;
|
||||
const usedExpected = new Set<number>();
|
||||
|
||||
for (const tx of extracted) {
|
||||
for (let i = 0; i < expected.length; i++) {
|
||||
if (usedExpected.has(i)) continue;
|
||||
|
||||
const exp = expected[i];
|
||||
const dateMatch = tx.date === exp.date;
|
||||
const amountMatch = Math.abs(tx.amount - exp.amount) < 0.02;
|
||||
|
||||
if (dateMatch && amountMatch) {
|
||||
matched++;
|
||||
usedExpected.add(i);
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return {
|
||||
matched,
|
||||
total: expected.length,
|
||||
accuracy: expected.length > 0 ? (matched / expected.length) * 100 : 0,
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
* Find all test cases (PDF + JSON pairs) in .nogit/bankstatements/
|
||||
*/
|
||||
function findTestCases(): Array<{ name: string; pdfPath: string; jsonPath: string }> {
|
||||
const testDir = path.join(process.cwd(), '.nogit/bankstatements');
|
||||
if (!fs.existsSync(testDir)) {
|
||||
return [];
|
||||
}
|
||||
|
||||
const files = fs.readdirSync(testDir);
|
||||
const pdfFiles = files.filter((f) => f.endsWith('.pdf'));
|
||||
const testCases: Array<{ name: string; pdfPath: string; jsonPath: string }> = [];
|
||||
|
||||
for (const pdf of pdfFiles) {
|
||||
const baseName = pdf.replace('.pdf', '');
|
||||
const jsonFile = `${baseName}.json`;
|
||||
if (files.includes(jsonFile)) {
|
||||
testCases.push({
|
||||
name: baseName,
|
||||
pdfPath: path.join(testDir, pdf),
|
||||
jsonPath: path.join(testDir, jsonFile),
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
testCases.sort((a, b) => a.name.localeCompare(b.name));
|
||||
return testCases;
|
||||
}
|
||||
|
||||
// Tests
|
||||
|
||||
tap.test('setup: ensure Docker containers are running', async () => {
|
||||
console.log('\n[Setup] Checking Docker containers...\n');
|
||||
|
||||
// Ensure PaddleOCR-VL Full Pipeline is running
|
||||
const paddleOk = await ensurePaddleOcrVlFull();
|
||||
expect(paddleOk).toBeTrue();
|
||||
|
||||
// Ensure MiniCPM is running (for field extraction from Markdown)
|
||||
const minicpmOk = await ensureMiniCpm();
|
||||
expect(minicpmOk).toBeTrue();
|
||||
|
||||
console.log('\n[Setup] All containers ready!\n');
|
||||
});
|
||||
|
||||
// Dynamic test for each PDF/JSON pair
|
||||
const testCases = findTestCases();
|
||||
console.log(`\nFound ${testCases.length} bank statement test cases (PaddleOCR-VL Full Pipeline)\n`);
|
||||
|
||||
const results: Array<{ name: string; accuracy: number; matched: number; total: number }> = [];
|
||||
|
||||
for (const testCase of testCases) {
|
||||
tap.test(`should extract bank statement: ${testCase.name}`, async () => {
|
||||
// Load expected data
|
||||
const expected: ITransaction[] = JSON.parse(fs.readFileSync(testCase.jsonPath, 'utf-8'));
|
||||
console.log(`\n=== ${testCase.name} ===`);
|
||||
console.log(`Expected: ${expected.length} transactions`);
|
||||
|
||||
const startTime = Date.now();
|
||||
|
||||
// Convert PDF to images
|
||||
const images = convertPdfToImages(testCase.pdfPath);
|
||||
console.log(` Pages: ${images.length}`);
|
||||
|
||||
// Extract all transactions
|
||||
const extracted = await extractAllTransactions(images);
|
||||
|
||||
const endTime = Date.now();
|
||||
const elapsedMs = endTime - startTime;
|
||||
|
||||
// Calculate accuracy
|
||||
const accuracy = calculateAccuracy(extracted, expected);
|
||||
results.push({
|
||||
name: testCase.name,
|
||||
accuracy: accuracy.accuracy,
|
||||
matched: accuracy.matched,
|
||||
total: accuracy.total,
|
||||
});
|
||||
|
||||
console.log(` Extracted: ${extracted.length} transactions`);
|
||||
console.log(` Matched: ${accuracy.matched}/${accuracy.total} (${accuracy.accuracy.toFixed(1)}%)`);
|
||||
console.log(` Time: ${(elapsedMs / 1000).toFixed(1)}s`);
|
||||
|
||||
// We expect at least 50% accuracy
|
||||
expect(accuracy.accuracy).toBeGreaterThan(50);
|
||||
});
|
||||
}
|
||||
|
||||
tap.test('summary', async () => {
|
||||
const totalStatements = results.length;
|
||||
const avgAccuracy =
|
||||
results.length > 0 ? results.reduce((a, b) => a + b.accuracy, 0) / results.length : 0;
|
||||
const totalMatched = results.reduce((a, b) => a + b.matched, 0);
|
||||
const totalExpected = results.reduce((a, b) => a + b.total, 0);
|
||||
|
||||
console.log(`\n======================================================`);
|
||||
console.log(` Bank Statement Extraction Summary (PaddleOCR-VL Full)`);
|
||||
console.log(`======================================================`);
|
||||
console.log(` Method: PaddleOCR-VL Full Pipeline -> MiniCPM`);
|
||||
console.log(` Statements: ${totalStatements}`);
|
||||
console.log(` Transactions: ${totalMatched}/${totalExpected} matched`);
|
||||
console.log(` Avg accuracy: ${avgAccuracy.toFixed(1)}%`);
|
||||
console.log(`======================================================\n`);
|
||||
});
|
||||
|
||||
export default tap.start();
|
||||
@@ -1,11 +1,19 @@
|
||||
/**
|
||||
* Invoice extraction test using MiniCPM-V (visual) + PaddleOCR-VL (OCR augmentation)
|
||||
*
|
||||
* This is the combined approach that uses both models for best accuracy:
|
||||
* - MiniCPM-V for visual understanding
|
||||
* - PaddleOCR-VL for OCR text to augment prompts
|
||||
*/
|
||||
import { tap, expect } from '@git.zone/tstest/tapbundle';
|
||||
import * as fs from 'fs';
|
||||
import * as path from 'path';
|
||||
import { execSync } from 'child_process';
|
||||
import * as os from 'os';
|
||||
import { ensurePaddleOcrVl, ensureMiniCpm } from './helpers/docker.js';
|
||||
|
||||
const OLLAMA_URL = 'http://localhost:11434';
|
||||
const MODEL = 'openbmb/minicpm-v4.5:q8_0';
|
||||
const MODEL = 'minicpm-v:latest';
|
||||
const PADDLEOCR_VL_URL = 'http://localhost:8000';
|
||||
|
||||
interface IInvoice {
|
||||
@@ -358,11 +366,18 @@ function findTestCases(): Array<{ name: string; pdfPath: string; jsonPath: strin
|
||||
|
||||
// Tests
|
||||
|
||||
tap.test('should connect to Ollama API', async () => {
|
||||
const response = await fetch(`${OLLAMA_URL}/api/tags`);
|
||||
expect(response.ok).toBeTrue();
|
||||
const data = await response.json();
|
||||
expect(data.models).toBeArray();
|
||||
tap.test('setup: ensure Docker containers are running', async () => {
|
||||
console.log('\n[Setup] Checking Docker containers...\n');
|
||||
|
||||
// Ensure PaddleOCR-VL is running (auto-detects GPU/CPU)
|
||||
const paddleOk = await ensurePaddleOcrVl();
|
||||
expect(paddleOk).toBeTrue();
|
||||
|
||||
// Ensure MiniCPM is running
|
||||
const minicpmOk = await ensureMiniCpm();
|
||||
expect(minicpmOk).toBeTrue();
|
||||
|
||||
console.log('\n[Setup] All containers ready!\n');
|
||||
});
|
||||
|
||||
tap.test('should have MiniCPM-V 4.5 model loaded', async () => {
|
||||
|
||||
345
test/test.invoices.minicpm.ts
Normal file
345
test/test.invoices.minicpm.ts
Normal file
@@ -0,0 +1,345 @@
|
||||
/**
|
||||
* Invoice extraction test using MiniCPM-V only (visual extraction)
|
||||
*
|
||||
* This tests MiniCPM-V's ability to extract invoice data directly from images
|
||||
* without any OCR augmentation.
|
||||
*/
|
||||
import { tap, expect } from '@git.zone/tstest/tapbundle';
|
||||
import * as fs from 'fs';
|
||||
import * as path from 'path';
|
||||
import { execSync } from 'child_process';
|
||||
import * as os from 'os';
|
||||
import { ensureMiniCpm } from './helpers/docker.js';
|
||||
|
||||
const OLLAMA_URL = 'http://localhost:11434';
|
||||
const MODEL = 'minicpm-v:latest';
|
||||
|
||||
interface IInvoice {
|
||||
invoice_number: string;
|
||||
invoice_date: string;
|
||||
vendor_name: string;
|
||||
currency: string;
|
||||
net_amount: number;
|
||||
vat_amount: number;
|
||||
total_amount: number;
|
||||
}
|
||||
|
||||
/**
|
||||
* Build extraction prompt (MiniCPM-V only, no OCR augmentation)
|
||||
*/
|
||||
function buildPrompt(): string {
|
||||
return `/nothink
|
||||
You are an invoice parser. Extract the following fields from this invoice:
|
||||
|
||||
1. invoice_number: The invoice/receipt number
|
||||
2. invoice_date: Date in YYYY-MM-DD format
|
||||
3. vendor_name: Company that issued the invoice
|
||||
4. currency: EUR, USD, etc.
|
||||
5. net_amount: Amount before tax (if shown)
|
||||
6. vat_amount: Tax/VAT amount (if shown, 0 if reverse charge or no tax)
|
||||
7. total_amount: Final amount due
|
||||
|
||||
Return ONLY valid JSON in this exact format:
|
||||
{"invoice_number":"XXX","invoice_date":"YYYY-MM-DD","vendor_name":"Company Name","currency":"EUR","net_amount":100.00,"vat_amount":19.00,"total_amount":119.00}
|
||||
|
||||
If a field is not visible, use null for strings or 0 for numbers.
|
||||
No explanation, just the JSON object.`;
|
||||
}
|
||||
|
||||
/**
|
||||
* Convert PDF to PNG images using ImageMagick
|
||||
*/
|
||||
function convertPdfToImages(pdfPath: string): string[] {
|
||||
const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'pdf-convert-'));
|
||||
const outputPattern = path.join(tempDir, 'page-%d.png');
|
||||
|
||||
try {
|
||||
execSync(
|
||||
`convert -density 200 -quality 90 "${pdfPath}" -background white -alpha remove "${outputPattern}"`,
|
||||
{ stdio: 'pipe' }
|
||||
);
|
||||
|
||||
const files = fs.readdirSync(tempDir).filter((f) => f.endsWith('.png')).sort();
|
||||
const images: string[] = [];
|
||||
|
||||
for (const file of files) {
|
||||
const imagePath = path.join(tempDir, file);
|
||||
const imageData = fs.readFileSync(imagePath);
|
||||
images.push(imageData.toString('base64'));
|
||||
}
|
||||
|
||||
return images;
|
||||
} finally {
|
||||
fs.rmSync(tempDir, { recursive: true, force: true });
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Single extraction pass with MiniCPM-V
|
||||
*/
|
||||
async function extractOnce(images: string[], passNum: number): Promise<IInvoice> {
|
||||
const payload = {
|
||||
model: MODEL,
|
||||
prompt: buildPrompt(),
|
||||
images,
|
||||
stream: true,
|
||||
options: {
|
||||
num_predict: 2048,
|
||||
temperature: 0.1,
|
||||
},
|
||||
};
|
||||
|
||||
const response = await fetch(`${OLLAMA_URL}/api/generate`, {
|
||||
method: 'POST',
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
body: JSON.stringify(payload),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
throw new Error(`Ollama API error: ${response.status}`);
|
||||
}
|
||||
|
||||
const reader = response.body?.getReader();
|
||||
if (!reader) {
|
||||
throw new Error('No response body');
|
||||
}
|
||||
|
||||
const decoder = new TextDecoder();
|
||||
let fullText = '';
|
||||
|
||||
while (true) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) break;
|
||||
|
||||
const chunk = decoder.decode(value, { stream: true });
|
||||
const lines = chunk.split('\n').filter((l) => l.trim());
|
||||
|
||||
for (const line of lines) {
|
||||
try {
|
||||
const json = JSON.parse(line);
|
||||
if (json.response) {
|
||||
fullText += json.response;
|
||||
}
|
||||
} catch {
|
||||
// Skip invalid JSON lines
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Extract JSON from response
|
||||
const startIdx = fullText.indexOf('{');
|
||||
const endIdx = fullText.lastIndexOf('}') + 1;
|
||||
|
||||
if (startIdx < 0 || endIdx <= startIdx) {
|
||||
throw new Error(`No JSON object found in response: ${fullText.substring(0, 200)}`);
|
||||
}
|
||||
|
||||
const jsonStr = fullText.substring(startIdx, endIdx);
|
||||
return JSON.parse(jsonStr);
|
||||
}
|
||||
|
||||
/**
|
||||
* Create a hash of invoice for comparison (using key fields)
|
||||
*/
|
||||
function hashInvoice(invoice: IInvoice): string {
|
||||
return `${invoice.invoice_number}|${invoice.invoice_date}|${invoice.total_amount.toFixed(2)}`;
|
||||
}
|
||||
|
||||
/**
|
||||
* Extract with consensus voting using MiniCPM-V only
|
||||
*/
|
||||
async function extractWithConsensus(images: string[], invoiceName: string, maxPasses: number = 5): Promise<IInvoice> {
|
||||
const results: Array<{ invoice: IInvoice; hash: string }> = [];
|
||||
const hashCounts: Map<string, number> = new Map();
|
||||
|
||||
const addResult = (invoice: IInvoice, passLabel: string): number => {
|
||||
const hash = hashInvoice(invoice);
|
||||
results.push({ invoice, hash });
|
||||
hashCounts.set(hash, (hashCounts.get(hash) || 0) + 1);
|
||||
console.log(` [${passLabel}] ${invoice.invoice_number} | ${invoice.invoice_date} | ${invoice.total_amount} ${invoice.currency}`);
|
||||
return hashCounts.get(hash)!;
|
||||
};
|
||||
|
||||
for (let pass = 1; pass <= maxPasses; pass++) {
|
||||
try {
|
||||
const invoice = await extractOnce(images, pass);
|
||||
const count = addResult(invoice, `Pass ${pass}`);
|
||||
|
||||
if (count >= 2) {
|
||||
console.log(` [Consensus] Reached after ${pass} passes`);
|
||||
return invoice;
|
||||
}
|
||||
} catch (err) {
|
||||
console.log(` [Pass ${pass}] Error: ${err}`);
|
||||
}
|
||||
}
|
||||
|
||||
// No consensus reached - return the most common result
|
||||
let bestHash = '';
|
||||
let bestCount = 0;
|
||||
for (const [hash, count] of hashCounts) {
|
||||
if (count > bestCount) {
|
||||
bestCount = count;
|
||||
bestHash = hash;
|
||||
}
|
||||
}
|
||||
|
||||
if (!bestHash) {
|
||||
throw new Error(`No valid results for ${invoiceName}`);
|
||||
}
|
||||
|
||||
const best = results.find((r) => r.hash === bestHash)!;
|
||||
console.log(` [No consensus] Using most common result (${bestCount}/${maxPasses} passes)`);
|
||||
return best.invoice;
|
||||
}
|
||||
|
||||
/**
|
||||
* Compare extracted invoice against expected
|
||||
*/
|
||||
function compareInvoice(
|
||||
extracted: IInvoice,
|
||||
expected: IInvoice
|
||||
): { match: boolean; errors: string[] } {
|
||||
const errors: string[] = [];
|
||||
|
||||
// Compare invoice number (normalize by removing spaces and case)
|
||||
const extNum = extracted.invoice_number?.replace(/\s/g, '').toLowerCase() || '';
|
||||
const expNum = expected.invoice_number?.replace(/\s/g, '').toLowerCase() || '';
|
||||
if (extNum !== expNum) {
|
||||
errors.push(`invoice_number: expected "${expected.invoice_number}", got "${extracted.invoice_number}"`);
|
||||
}
|
||||
|
||||
// Compare date
|
||||
if (extracted.invoice_date !== expected.invoice_date) {
|
||||
errors.push(`invoice_date: expected "${expected.invoice_date}", got "${extracted.invoice_date}"`);
|
||||
}
|
||||
|
||||
// Compare total amount (with tolerance)
|
||||
if (Math.abs(extracted.total_amount - expected.total_amount) > 0.02) {
|
||||
errors.push(`total_amount: expected ${expected.total_amount}, got ${extracted.total_amount}`);
|
||||
}
|
||||
|
||||
// Compare currency
|
||||
if (extracted.currency?.toUpperCase() !== expected.currency?.toUpperCase()) {
|
||||
errors.push(`currency: expected "${expected.currency}", got "${extracted.currency}"`);
|
||||
}
|
||||
|
||||
return { match: errors.length === 0, errors };
|
||||
}
|
||||
|
||||
/**
|
||||
* Find all test cases (PDF + JSON pairs) in .nogit/invoices/
|
||||
*/
|
||||
function findTestCases(): Array<{ name: string; pdfPath: string; jsonPath: string }> {
|
||||
const testDir = path.join(process.cwd(), '.nogit/invoices');
|
||||
if (!fs.existsSync(testDir)) {
|
||||
return [];
|
||||
}
|
||||
|
||||
const files = fs.readdirSync(testDir);
|
||||
const pdfFiles = files.filter((f) => f.endsWith('.pdf'));
|
||||
const testCases: Array<{ name: string; pdfPath: string; jsonPath: string }> = [];
|
||||
|
||||
for (const pdf of pdfFiles) {
|
||||
const baseName = pdf.replace('.pdf', '');
|
||||
const jsonFile = `${baseName}.json`;
|
||||
if (files.includes(jsonFile)) {
|
||||
testCases.push({
|
||||
name: baseName,
|
||||
pdfPath: path.join(testDir, pdf),
|
||||
jsonPath: path.join(testDir, jsonFile),
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
// Sort alphabetically
|
||||
testCases.sort((a, b) => a.name.localeCompare(b.name));
|
||||
|
||||
return testCases;
|
||||
}
|
||||
|
||||
// Tests
|
||||
|
||||
tap.test('setup: ensure Docker containers are running', async () => {
|
||||
console.log('\n[Setup] Checking Docker containers...\n');
|
||||
|
||||
// Ensure MiniCPM is running
|
||||
const minicpmOk = await ensureMiniCpm();
|
||||
expect(minicpmOk).toBeTrue();
|
||||
|
||||
console.log('\n[Setup] All containers ready!\n');
|
||||
});
|
||||
|
||||
tap.test('should have MiniCPM-V 4.5 model loaded', async () => {
|
||||
const response = await fetch(`${OLLAMA_URL}/api/tags`);
|
||||
const data = await response.json();
|
||||
const modelNames = data.models.map((m: { name: string }) => m.name);
|
||||
expect(modelNames.some((name: string) => name.includes('minicpm-v4.5'))).toBeTrue();
|
||||
});
|
||||
|
||||
// Dynamic test for each PDF/JSON pair
|
||||
const testCases = findTestCases();
|
||||
console.log(`\nFound ${testCases.length} invoice test cases (MiniCPM-V only)\n`);
|
||||
|
||||
let passedCount = 0;
|
||||
let failedCount = 0;
|
||||
const processingTimes: number[] = [];
|
||||
|
||||
for (const testCase of testCases) {
|
||||
tap.test(`should extract invoice: ${testCase.name}`, async () => {
|
||||
// Load expected data
|
||||
const expected: IInvoice = JSON.parse(fs.readFileSync(testCase.jsonPath, 'utf-8'));
|
||||
console.log(`\n=== ${testCase.name} ===`);
|
||||
console.log(`Expected: ${expected.invoice_number} | ${expected.invoice_date} | ${expected.total_amount} ${expected.currency}`);
|
||||
|
||||
const startTime = Date.now();
|
||||
|
||||
// Convert PDF to images
|
||||
const images = convertPdfToImages(testCase.pdfPath);
|
||||
console.log(` Pages: ${images.length}`);
|
||||
|
||||
// Extract with consensus voting (MiniCPM-V only)
|
||||
const extracted = await extractWithConsensus(images, testCase.name);
|
||||
|
||||
const endTime = Date.now();
|
||||
const elapsedMs = endTime - startTime;
|
||||
processingTimes.push(elapsedMs);
|
||||
|
||||
// Compare results
|
||||
const result = compareInvoice(extracted, expected);
|
||||
|
||||
if (result.match) {
|
||||
passedCount++;
|
||||
console.log(` Result: MATCH (${(elapsedMs / 1000).toFixed(1)}s)`);
|
||||
} else {
|
||||
failedCount++;
|
||||
console.log(` Result: MISMATCH (${(elapsedMs / 1000).toFixed(1)}s)`);
|
||||
result.errors.forEach((e) => console.log(` - ${e}`));
|
||||
}
|
||||
|
||||
// Assert match
|
||||
expect(result.match).toBeTrue();
|
||||
});
|
||||
}
|
||||
|
||||
tap.test('summary', async () => {
|
||||
const totalInvoices = testCases.length;
|
||||
const accuracy = totalInvoices > 0 ? (passedCount / totalInvoices) * 100 : 0;
|
||||
const totalTimeMs = processingTimes.reduce((a, b) => a + b, 0);
|
||||
const avgTimeMs = processingTimes.length > 0 ? totalTimeMs / processingTimes.length : 0;
|
||||
const avgTimeSec = avgTimeMs / 1000;
|
||||
const totalTimeSec = totalTimeMs / 1000;
|
||||
|
||||
console.log(`\n========================================`);
|
||||
console.log(` Invoice Extraction Summary (MiniCPM)`);
|
||||
console.log(`========================================`);
|
||||
console.log(` Passed: ${passedCount}/${totalInvoices}`);
|
||||
console.log(` Failed: ${failedCount}/${totalInvoices}`);
|
||||
console.log(` Accuracy: ${accuracy.toFixed(1)}%`);
|
||||
console.log(`----------------------------------------`);
|
||||
console.log(` Total time: ${totalTimeSec.toFixed(1)}s`);
|
||||
console.log(` Avg per inv: ${avgTimeSec.toFixed(1)}s`);
|
||||
console.log(`========================================\n`);
|
||||
});
|
||||
|
||||
export default tap.start();
|
||||
393
test/test.invoices.paddleocr-vl.ts
Normal file
393
test/test.invoices.paddleocr-vl.ts
Normal file
@@ -0,0 +1,393 @@
|
||||
/**
|
||||
* Invoice extraction test using PaddleOCR-VL Full Pipeline
|
||||
*
|
||||
* This tests the complete PaddleOCR-VL pipeline:
|
||||
* 1. PP-DocLayoutV2 for layout detection
|
||||
* 2. PaddleOCR-VL for recognition
|
||||
* 3. Structured Markdown output
|
||||
* 4. MiniCPM extracts invoice fields from structured Markdown
|
||||
*
|
||||
* The structured Markdown has proper tables and formatting,
|
||||
* making it much easier for MiniCPM to extract invoice data.
|
||||
*/
|
||||
import { tap, expect } from '@git.zone/tstest/tapbundle';
|
||||
import * as fs from 'fs';
|
||||
import * as path from 'path';
|
||||
import { execSync } from 'child_process';
|
||||
import * as os from 'os';
|
||||
import { ensurePaddleOcrVlFull, ensureMiniCpm } from './helpers/docker.js';
|
||||
|
||||
const PADDLEOCR_VL_URL = 'http://localhost:8000';
|
||||
const OLLAMA_URL = 'http://localhost:11434';
|
||||
const MINICPM_MODEL = 'minicpm-v:latest';
|
||||
|
||||
interface IInvoice {
|
||||
invoice_number: string;
|
||||
invoice_date: string;
|
||||
vendor_name: string;
|
||||
currency: string;
|
||||
net_amount: number;
|
||||
vat_amount: number;
|
||||
total_amount: number;
|
||||
}
|
||||
|
||||
/**
|
||||
* Convert PDF to PNG images using ImageMagick
|
||||
*/
|
||||
function convertPdfToImages(pdfPath: string): string[] {
|
||||
const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'pdf-convert-'));
|
||||
const outputPattern = path.join(tempDir, 'page-%d.png');
|
||||
|
||||
try {
|
||||
execSync(
|
||||
`convert -density 200 -quality 90 "${pdfPath}" -background white -alpha remove "${outputPattern}"`,
|
||||
{ stdio: 'pipe' }
|
||||
);
|
||||
|
||||
const files = fs.readdirSync(tempDir).filter((f) => f.endsWith('.png')).sort();
|
||||
const images: string[] = [];
|
||||
|
||||
for (const file of files) {
|
||||
const imagePath = path.join(tempDir, file);
|
||||
const imageData = fs.readFileSync(imagePath);
|
||||
images.push(imageData.toString('base64'));
|
||||
}
|
||||
|
||||
return images;
|
||||
} finally {
|
||||
fs.rmSync(tempDir, { recursive: true, force: true });
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Parse document using PaddleOCR-VL Full Pipeline (returns structured Markdown)
|
||||
*/
|
||||
async function parseDocument(imageBase64: string): Promise<string> {
|
||||
const response = await fetch(`${PADDLEOCR_VL_URL}/parse`, {
|
||||
method: 'POST',
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
body: JSON.stringify({
|
||||
image: imageBase64,
|
||||
output_format: 'markdown',
|
||||
}),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
const text = await response.text();
|
||||
throw new Error(`PaddleOCR-VL API error: ${response.status} - ${text}`);
|
||||
}
|
||||
|
||||
const data = await response.json();
|
||||
|
||||
if (!data.success) {
|
||||
throw new Error(`PaddleOCR-VL error: ${data.error}`);
|
||||
}
|
||||
|
||||
return data.result?.markdown || '';
|
||||
}
|
||||
|
||||
/**
|
||||
* Extract invoice fields from structured Markdown using MiniCPM with image context
|
||||
*/
|
||||
async function extractInvoiceFromMarkdown(markdown: string, images: string[]): Promise<IInvoice> {
|
||||
// Truncate if too long
|
||||
const truncated = markdown.length > 8000 ? markdown.slice(0, 8000) : markdown;
|
||||
console.log(` [Extract] Processing ${truncated.length} chars of Markdown`);
|
||||
|
||||
const prompt = `/nothink
|
||||
You are an invoice parser. Extract fields from this invoice image.
|
||||
|
||||
Required fields:
|
||||
- invoice_number: The invoice/receipt number
|
||||
- invoice_date: Date in YYYY-MM-DD format
|
||||
- vendor_name: Company that issued the invoice
|
||||
- currency: EUR, USD, etc.
|
||||
- net_amount: Amount before tax
|
||||
- vat_amount: Tax/VAT amount (0 if reverse charge)
|
||||
- total_amount: Final amount due
|
||||
|
||||
Return ONLY a JSON object like:
|
||||
{"invoice_number":"123","invoice_date":"2022-01-28","vendor_name":"Adobe","currency":"EUR","net_amount":24.99,"vat_amount":0,"total_amount":24.99}
|
||||
|
||||
Use null for missing strings, 0 for missing numbers. No explanation.
|
||||
|
||||
OCR text from the invoice (for reference):
|
||||
---
|
||||
${truncated}
|
||||
---`;
|
||||
|
||||
const payload = {
|
||||
model: MINICPM_MODEL,
|
||||
prompt,
|
||||
images, // Send the actual image to MiniCPM
|
||||
stream: true,
|
||||
options: {
|
||||
num_predict: 2048,
|
||||
temperature: 0.1,
|
||||
},
|
||||
};
|
||||
|
||||
const response = await fetch(`${OLLAMA_URL}/api/generate`, {
|
||||
method: 'POST',
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
body: JSON.stringify(payload),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
throw new Error(`Ollama API error: ${response.status}`);
|
||||
}
|
||||
|
||||
const reader = response.body?.getReader();
|
||||
if (!reader) {
|
||||
throw new Error('No response body');
|
||||
}
|
||||
|
||||
const decoder = new TextDecoder();
|
||||
let fullText = '';
|
||||
|
||||
while (true) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) break;
|
||||
|
||||
const chunk = decoder.decode(value, { stream: true });
|
||||
const lines = chunk.split('\n').filter((l) => l.trim());
|
||||
|
||||
for (const line of lines) {
|
||||
try {
|
||||
const json = JSON.parse(line);
|
||||
if (json.response) {
|
||||
fullText += json.response;
|
||||
}
|
||||
} catch {
|
||||
// Skip invalid JSON lines
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Extract JSON from response
|
||||
const startIdx = fullText.indexOf('{');
|
||||
const endIdx = fullText.lastIndexOf('}') + 1;
|
||||
|
||||
if (startIdx < 0 || endIdx <= startIdx) {
|
||||
throw new Error(`No JSON object found in response: ${fullText.substring(0, 200)}`);
|
||||
}
|
||||
|
||||
const jsonStr = fullText.substring(startIdx, endIdx);
|
||||
return JSON.parse(jsonStr);
|
||||
}
|
||||
|
||||
/**
|
||||
* Single extraction pass: Parse with PaddleOCR-VL Full, extract with MiniCPM
|
||||
*/
|
||||
async function extractOnce(images: string[], passNum: number): Promise<IInvoice> {
|
||||
// Parse document with full pipeline
|
||||
const markdown = await parseDocument(images[0]);
|
||||
console.log(` [Parse] Got ${markdown.split('\n').length} lines of Markdown`);
|
||||
|
||||
// Extract invoice fields from Markdown with image context
|
||||
return extractInvoiceFromMarkdown(markdown, images);
|
||||
}
|
||||
|
||||
/**
|
||||
* Create a hash of invoice for comparison (using key fields)
|
||||
*/
|
||||
function hashInvoice(invoice: IInvoice): string {
|
||||
return `${invoice.invoice_number}|${invoice.invoice_date}|${invoice.total_amount.toFixed(2)}`;
|
||||
}
|
||||
|
||||
/**
|
||||
* Extract with consensus voting
|
||||
*/
|
||||
async function extractWithConsensus(images: string[], invoiceName: string, maxPasses: number = 5): Promise<IInvoice> {
|
||||
const results: Array<{ invoice: IInvoice; hash: string }> = [];
|
||||
const hashCounts: Map<string, number> = new Map();
|
||||
|
||||
const addResult = (invoice: IInvoice, passLabel: string): number => {
|
||||
const hash = hashInvoice(invoice);
|
||||
results.push({ invoice, hash });
|
||||
hashCounts.set(hash, (hashCounts.get(hash) || 0) + 1);
|
||||
console.log(` [${passLabel}] ${invoice.invoice_number} | ${invoice.invoice_date} | ${invoice.total_amount} ${invoice.currency}`);
|
||||
return hashCounts.get(hash)!;
|
||||
};
|
||||
|
||||
for (let pass = 1; pass <= maxPasses; pass++) {
|
||||
try {
|
||||
const invoice = await extractOnce(images, pass);
|
||||
const count = addResult(invoice, `Pass ${pass}`);
|
||||
|
||||
if (count >= 2) {
|
||||
console.log(` [Consensus] Reached after ${pass} passes`);
|
||||
return invoice;
|
||||
}
|
||||
} catch (err) {
|
||||
console.log(` [Pass ${pass}] Error: ${err}`);
|
||||
}
|
||||
}
|
||||
|
||||
// No consensus reached - return the most common result
|
||||
let bestHash = '';
|
||||
let bestCount = 0;
|
||||
for (const [hash, count] of hashCounts) {
|
||||
if (count > bestCount) {
|
||||
bestCount = count;
|
||||
bestHash = hash;
|
||||
}
|
||||
}
|
||||
|
||||
if (!bestHash) {
|
||||
throw new Error(`No valid results for ${invoiceName}`);
|
||||
}
|
||||
|
||||
const best = results.find((r) => r.hash === bestHash)!;
|
||||
console.log(` [No consensus] Using most common result (${bestCount}/${maxPasses} passes)`);
|
||||
return best.invoice;
|
||||
}
|
||||
|
||||
/**
|
||||
* Compare extracted invoice against expected
|
||||
*/
|
||||
function compareInvoice(
|
||||
extracted: IInvoice,
|
||||
expected: IInvoice
|
||||
): { match: boolean; errors: string[] } {
|
||||
const errors: string[] = [];
|
||||
|
||||
// Compare invoice number (normalize by removing spaces and case)
|
||||
const extNum = extracted.invoice_number?.replace(/\s/g, '').toLowerCase() || '';
|
||||
const expNum = expected.invoice_number?.replace(/\s/g, '').toLowerCase() || '';
|
||||
if (extNum !== expNum) {
|
||||
errors.push(`invoice_number: expected "${expected.invoice_number}", got "${extracted.invoice_number}"`);
|
||||
}
|
||||
|
||||
// Compare date
|
||||
if (extracted.invoice_date !== expected.invoice_date) {
|
||||
errors.push(`invoice_date: expected "${expected.invoice_date}", got "${extracted.invoice_date}"`);
|
||||
}
|
||||
|
||||
// Compare total amount (with tolerance)
|
||||
if (Math.abs(extracted.total_amount - expected.total_amount) > 0.02) {
|
||||
errors.push(`total_amount: expected ${expected.total_amount}, got ${extracted.total_amount}`);
|
||||
}
|
||||
|
||||
// Compare currency
|
||||
if (extracted.currency?.toUpperCase() !== expected.currency?.toUpperCase()) {
|
||||
errors.push(`currency: expected "${expected.currency}", got "${extracted.currency}"`);
|
||||
}
|
||||
|
||||
return { match: errors.length === 0, errors };
|
||||
}
|
||||
|
||||
/**
|
||||
* Find all test cases (PDF + JSON pairs) in .nogit/invoices/
|
||||
*/
|
||||
function findTestCases(): Array<{ name: string; pdfPath: string; jsonPath: string }> {
|
||||
const testDir = path.join(process.cwd(), '.nogit/invoices');
|
||||
if (!fs.existsSync(testDir)) {
|
||||
return [];
|
||||
}
|
||||
|
||||
const files = fs.readdirSync(testDir);
|
||||
const pdfFiles = files.filter((f) => f.endsWith('.pdf'));
|
||||
const testCases: Array<{ name: string; pdfPath: string; jsonPath: string }> = [];
|
||||
|
||||
for (const pdf of pdfFiles) {
|
||||
const baseName = pdf.replace('.pdf', '');
|
||||
const jsonFile = `${baseName}.json`;
|
||||
if (files.includes(jsonFile)) {
|
||||
testCases.push({
|
||||
name: baseName,
|
||||
pdfPath: path.join(testDir, pdf),
|
||||
jsonPath: path.join(testDir, jsonFile),
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
// Sort alphabetically
|
||||
testCases.sort((a, b) => a.name.localeCompare(b.name));
|
||||
|
||||
return testCases;
|
||||
}
|
||||
|
||||
// Tests
|
||||
|
||||
tap.test('setup: ensure Docker containers are running', async () => {
|
||||
console.log('\n[Setup] Checking Docker containers...\n');
|
||||
|
||||
// Ensure PaddleOCR-VL Full Pipeline is running
|
||||
const paddleOk = await ensurePaddleOcrVlFull();
|
||||
expect(paddleOk).toBeTrue();
|
||||
|
||||
// Ensure MiniCPM is running (for field extraction from Markdown)
|
||||
const minicpmOk = await ensureMiniCpm();
|
||||
expect(minicpmOk).toBeTrue();
|
||||
|
||||
console.log('\n[Setup] All containers ready!\n');
|
||||
});
|
||||
|
||||
// Dynamic test for each PDF/JSON pair
|
||||
const testCases = findTestCases();
|
||||
console.log(`\nFound ${testCases.length} invoice test cases (PaddleOCR-VL Full Pipeline)\n`);
|
||||
|
||||
let passedCount = 0;
|
||||
let failedCount = 0;
|
||||
const processingTimes: number[] = [];
|
||||
|
||||
for (const testCase of testCases) {
|
||||
tap.test(`should extract invoice: ${testCase.name}`, async () => {
|
||||
// Load expected data
|
||||
const expected: IInvoice = JSON.parse(fs.readFileSync(testCase.jsonPath, 'utf-8'));
|
||||
console.log(`\n=== ${testCase.name} ===`);
|
||||
console.log(`Expected: ${expected.invoice_number} | ${expected.invoice_date} | ${expected.total_amount} ${expected.currency}`);
|
||||
|
||||
const startTime = Date.now();
|
||||
|
||||
// Convert PDF to images
|
||||
const images = convertPdfToImages(testCase.pdfPath);
|
||||
console.log(` Pages: ${images.length}`);
|
||||
|
||||
// Extract with consensus voting (PaddleOCR-VL Full -> MiniCPM)
|
||||
const extracted = await extractWithConsensus(images, testCase.name);
|
||||
|
||||
const endTime = Date.now();
|
||||
const elapsedMs = endTime - startTime;
|
||||
processingTimes.push(elapsedMs);
|
||||
|
||||
// Compare results
|
||||
const result = compareInvoice(extracted, expected);
|
||||
|
||||
if (result.match) {
|
||||
passedCount++;
|
||||
console.log(` Result: MATCH (${(elapsedMs / 1000).toFixed(1)}s)`);
|
||||
} else {
|
||||
failedCount++;
|
||||
console.log(` Result: MISMATCH (${(elapsedMs / 1000).toFixed(1)}s)`);
|
||||
result.errors.forEach((e) => console.log(` - ${e}`));
|
||||
}
|
||||
|
||||
// Assert match
|
||||
expect(result.match).toBeTrue();
|
||||
});
|
||||
}
|
||||
|
||||
tap.test('summary', async () => {
|
||||
const totalInvoices = testCases.length;
|
||||
const accuracy = totalInvoices > 0 ? (passedCount / totalInvoices) * 100 : 0;
|
||||
const totalTimeMs = processingTimes.reduce((a, b) => a + b, 0);
|
||||
const avgTimeMs = processingTimes.length > 0 ? totalTimeMs / processingTimes.length : 0;
|
||||
const avgTimeSec = avgTimeMs / 1000;
|
||||
const totalTimeSec = totalTimeMs / 1000;
|
||||
|
||||
console.log(`\n======================================================`);
|
||||
console.log(` Invoice Extraction Summary (PaddleOCR-VL Full)`);
|
||||
console.log(`======================================================`);
|
||||
console.log(` Method: PaddleOCR-VL Full Pipeline -> MiniCPM`);
|
||||
console.log(` Passed: ${passedCount}/${totalInvoices}`);
|
||||
console.log(` Failed: ${failedCount}/${totalInvoices}`);
|
||||
console.log(` Accuracy: ${accuracy.toFixed(1)}%`);
|
||||
console.log(`------------------------------------------------------`);
|
||||
console.log(` Total time: ${totalTimeSec.toFixed(1)}s`);
|
||||
console.log(` Avg per inv: ${avgTimeSec.toFixed(1)}s`);
|
||||
console.log(`======================================================\n`);
|
||||
});
|
||||
|
||||
export default tap.start();
|
||||
Reference in New Issue
Block a user