Unexpected Token

[imlanky]

/node_modules/upscaler/dist/browser/esm/index.js:1
export { default, } from './upscaler';
^^^^^^

SyntaxError: Unexpected token 'export'

[thekevinscott]

Can you provide a reproducible repository?

An issue like this is usually related to how your bundle or transpiler is set up. You're loading ESM code but whatever setup is complaining that it doesn't know how to handle it.

[thekevinscott]

Closing for now, feel free to reopen with a reproducible repo

[hollowshiroyuki]

Hello, reopening this issue with a reproducible repo https://github.com/hollowshiroyuki/Timi_OCR_Proto. I'm trying to increase TesseractJS results by upscaling before processing the picture. You need to uncomment lines 3, 6 and 17, and remove line 18 in index.ts.

Tested with :

• tsc v4.9.3
• node v19.2.0
• ts-node v10.9.2

Quick way to test : npm run run <Image File>

Thanks for your assistance !

[thekevinscott]

@hollowshiroyuki I can’t access that repo - maybe it’s set to private?

[hollowshiroyuki]

whoops sorry, it's fixed now 👍

[thekevinscott]

Thanks, I can see it now.

I’m on my phone and can check later, but at a glance - is this a Node.js project? If so you’ll need to import from node:

Import ‘upscaler/node’

check out the getting started docs section on node.

I’ll double check again when I’m front of my computer

[hollowshiroyuki]

This is indeed a nodejs project and I'm using Typescript.

I tried your suggestion but it didn't work, I'm not sure if it's a problem of the package or my typescript configuration.

I also tried using https://www.typescriptlang.org/docs/handbook/module-resolution.html#path-mapping without success.

{
  "compilerOptions": {
    "baseUrl": ".",
    "paths": {
      "upscaler": ["node_modules/upscaler/dist/node"],
    }
  }
}

[thekevinscott]

I opened https://github.com/hollowshiroyuki/Timi_OCR_Proto/pull/1/files which should resolve the issues you're seeing (I'm not familiar with TesseractJS, but the script completes without errors).

Some things:

• You need to install @tensorflow/tfjs-node as an explicit dependency alongside UpscalerJS
• You need to use require to load Upscaler in a Node environment - more on this below
• The output of an upscale operation is either a base64 string or a tensor. For your purposes (I think TesseractJS expects a file on disk) you're better off returning a tensor and saving it to disk. Base64 outputs are more useful in a browser experience for easy display in an <img /> tag.
• If you want to provide a custom scale, you'll also need to pass a path to a valid model.json file (alternatively you can import a specific model with something like this).

---

On the subject of needing to load the Node version with require: we bundle the library in a few flavors:

• ESM (browser only)
• CJS (node only)
• UMD (browser only)

Which has been roughly sufficient, as most Node projects can consume CJS (either because they're written in CJS themselves, or can use a transpiler) and most browser-based projects can consume either ESM (via a transpiler) or load the UMD script tag.

It gets a bit trickier with the use of a tool like ts-node, which, as far as I understand, is transpiling your ESM Node source code on the fly with tsc into CommonJS, but which is expecting imported libraries to also be ESM.

The reason I haven't provided an ESM-compatible Node version of the library so far is that Node also supports native ESM, but via the use of .mjs extensions, which (in the limited investigation I did) wrecked havoc with the existing build process.

That said, this is good motivation to pick that investigation back up, as you probably are not the only one who will run into that issue (and it's weird to need a single require() tag amongst the other import tags).

[hollowshiroyuki]

Indeed it works now, we got better recognition performance, thanks a lot !

[thekevinscott]

Awesome, great to hear!

One other thing I should have mentioned, because you're in Node, you should experiment with providing an explicit model. I'd recommend either the 4x or 8x models from esrgan-thick (the 8x image is not shown, which is a bug in the documentation, but you should be able to import it). You can also explicitly specify the model you're using now, GANS from esrgan-legacy.

The reason I recommend this is because UpscalerJS is using the default-model by default, which currently is the GANS model (slow and more precise) but which will be switching to the 2x esrgan-slim model (fast, less precise). For your use case, I think you probably want the most precise model.