Edit this page

handwritten-score-recognition-0003

Use Case and High-Level Description

This is a network for text recognition scenario. It consists of VGG16-like backbone and bidirectional LSTM encoder-decoder. The network is able to recognize school marks that should have format either <digit> or <digit>.<digit> (e.g. 4 or 3.5).

Example

_images/shot_25.png

-> Mark2.5

Specification

Metric

Value

Accuracy (internal test set)

98.83%

Text location requirements

Tight aligned crop

GFlops

0.792

MParams

5.555

Source framework

TensorFlow*

Inputs

Image, name: Placeholder, shape: 1, 1, 32, 64 in the format B, C, H, W, where:

  • B - batch size

  • C - number of channels

  • H - image height

  • W - image width

Note that the source image should be tight aligned crop with detected text converted to grayscale.

Outputs

The net outputs a blob with the shape 16, 1, 13 in the format W, B, L, where:

  • W - output sequence length

  • B - batch size

  • L - confidence distribution across the alphabet: "0123456789._#", where # - special blank character for CTC decoding algorithm and the character _’` replaces all non-numeric symbols.

The network output can be decoded by CTC Greedy Decoder or CTC Beam Search decoder.

Prev Next