[Fix] PyTorch MASTER implementation by felixdittrich92 · Pull Request #941 · mindee/doctr (original) (raw)

This PR: (Still in progress)

• fix decoding step
• fix Decoder

Any feedback is welcome 🤗

some notes at the moment:
different from paper input set to our default (32, 128)

ftm: transformer plain implementation not nn modules

but training and inference works fine 😅

@frgfm feel free to grab this and take a look 👍

Issue:
#802