在 Google Cloud Vision、Microsoft Azure Computer Vision API 或 Amazon Text Extract 上进行微调

Question

I need to transcribe a large number of Handwritten documents. I tried to use cloud services from either Google, Amazon, and Microsoft. Namely:

• https://azure.microsoft.com/en-us/services/cognitive-services/computer-vision/
• https://cloud.google.com/vision/docs/handwriting
• https://aws.amazon.com/textract/

Unfortunately, none of them achieved good enough results. I suspect it is because my documents have a weird handwriting style, and as a result, the networks struggle a lot. I searched whether I could fine-tune (with manually transcribed data), but I have not found anything online, so as a last resort, I ask here.

If it is possible to fine-tune one of these models, could you please point me some resources?

Answer 1

You are correct, with Azure Cognitive Services with Computer Vision you cannot upload your own data to train the API to recognise the handwriting in your documents I'm afraid. I can't comment on the other offerings from AWS and Google I'm afraid, but certainly not for Azure.