This issue is to be used to track the roadmap of docTR for release v0.5.1, and collect feedback from users & contributors.

• Datasets

  • Add SynthWordGenerator to text reco training scripts #825
  • [datasets] Targets are modified inplace #840

• Models

  • Bridge the gap between TensorFlow & PyTorch support of text detection on rotation (feat: add rotated linknet_resnet18 tensorflow ckpts #817)
  • Restore remap_boxes function #800

• Transforms

  • Harmonize RandomRotate between PyTorch & TensorFlow fix: Fixed rotation of img + target #784
  • Add missing TensorFlow support for docTR transforms feat: Added RandomHorizontalFLip in TF #779

• PDF

  • Consider removing PyMuPDF for dependency that is not AGPL licensed #486

• Documentation

  • [docs] Document usage of predictors on skewed documents #807

• References

  • Cannot train pytorch sar_resnet31 and master recognition model #802