ML Number Recognizer Model
Pre-release
Pre-release
·
2 commits
to main
since this release
Trained on RTX3060 for 50 Epochs
The synthetic dataset was generated using MNIST
Maximum Number length kept at 5
Trained up to number 20000
The model will output a [1,32,13] Tensor on which the CTC decoder must be applied.
The Tensorflow model contains the entire model.
For onnx the inference code is in the Notebooks directory
The TF SavedModel format is still WIP