Service Centre Reviews

“struggles with accuracy on complex documents

Pros

The best thing about it is how much control it gives you. Even when the base model produces less accurate OCR output, you can improve it by tweaking the parameters. There are tons of configuration options available and if your data follows certain patterns, these settings really come in handy for boosting results.

Cons

The accuracy isn't as good as other commercial OCR solutions I've tried. When image quality varies or there's lots of numbers and fractions, getting reliable results becomes really challenging. Also, while there are a lot of parameters you can tweak, the documentation doesn't make it clear how to actually use them effectively.

“Very accurate

Pros

OCR accuracy is really impressive, especially when I preprocess images with grayscale first. Response times are quite fast too.

Cons

I feel like it struggles with non-text elements in screenshots like when I tried to capture Chrome's disconnect page, it mistakenly converted the dinosaur icon into the number 10. You'll need to clean up the inputs afterwards to get accurate results.

“Impressed by how accurate it is

Pros

I like how accurate it is and the PSM feature definitely takes the extraction precision to another level.

Cons

Sometimes it fails to pull information correctly from handwritten documents.

“Not there yet!

Pros

Setting this up was simple and didn't take much time to get running. Plus, having Google behind the maintenance gives me confidence in its reliability.

Cons

The performance is pretty disappointing across most scenarios I encounter. There's no handwritten text support, which is really essential nowadays. Even some research models trained on limited datasets outperform it for scene text detection tasks.

“Difficult to get settings right

Pros

I can pull text from documents in multiple languages without any difficulty which saves me time when dealing with international content.

Cons

Getting the settings right takes way too many tries. Meanwhile, Microsoft Cognitive OCR and Textract make it much simpler to achieve good results without all the back-and-forth tweaking.