aster
aster copied to clipboard
Multi detection
Is it possible to detect and recognise multiple texts from a scene using aster?
Yes. Predict 1 text box for each word that appears in the image and then predict the contents of each textbox. Where in the files should i look to be able to do that?
Nevermind, I might have been wrong, apologies.
I found something that may hopefully help you.
In aster/protos/train.proto
the 8th defined field around line 35 called from_detection_checkpoint
is an optional bool that allows you to specify if your fine_tune_checkpoint
was from an object detection model.
This appears to default to False
, meaning the default training configuration assumes that you're using a CNN instead of an R-CNN.
Hopefully you're able to configure this and make it work. If you find a pre-trained detection model that ends up working, please let us know.