Closed amitdo closed 2 hours ago
It's only remaining use for me is that the code shows how to add another OCR engine, for example to use Kraken models with optional GPU support (on my wish list).
IIRC, the code in Tesseract uses some internal TF 1.x functionality instead of a stable API.
I think I can look in old releases how an OCR engine was added, so we can indeed remove the TensorFlow code now.
All the files under src/lstm that stars with
tf
.It is useless. There is no way to train a model in TF that Tesseract can use for inference.
Can we finally remove the TF code from Tesseract?