text-recognition-0016 (composite)

Use Case and High-Level Description

This is a text-recognition composite model that recognizes scene text. The model uses predefined set of alphanumeric symbols (case-insensitive) to predict words. The model is built on the ResNeXt-101 backbone with TPS module and additional 2d attention-based text recognition head.

Example of the input data

Example of the output

openvino

Composite model specification

Metric

Value

Text location requirements

Tight aligned crop

Source framework

PyTorch*

Accuracy on the alphanumeric subset of ICDAR13

0.9685

Accuracy on the alphanumeric subset of ICDAR03

0.9712

Accuracy on the alphanumeric subset of ICDAR15

0.8675

Accuracy on the alphanumeric subset of SVT

0.9474

Accuracy on the alphanumeric subset of IIIT5K

0.9347

Encoder model specification

The text-recognition-0016-encoder model is a ResNeXt-101 like backbone with TPS network and convolutional encoder part of the text recognition.

Metric

Value

GFlops

9.27

MParams

88.1

Inputs

Image, name: imgs, shape: 1, 1, 64, 256 in the B, C, H, W format, where:

  • B - batch size

  • C - number of channels

  • H - image height

  • W - image width

Outputs

  1. Name: decoder_hidden, shape: 1, 1, 1024. Initial context state of the GRU cell.

  2. Name: features, shape: 1, 36, 1024. Features from encoder part of text recognition head.

Decoder model specification

The text-recognition-0016-decoder model is a GRU based decoder with 2d attention module.

Metric

Value

GFlops

0.08

MParams

4.28

Inputs

  1. Name: decoder_input, shape: 1. Previous predicted letter.

  2. Name: features, shape: 1, 36, 1024. Encoded features.

  3. Name: hidden, shape: 1, 1, 1024. Current state of the decoder.

Outputs

  1. Name: decoder_hidden, shape: 1, 1, 1024. Current context state of the GRU cell.

  2. Name: decoder_output, shape: 1, 40. Classification confidence scores in the [0, 1] range for every letter.

Particularly, decoder output in every step is the probability distribution of the symbol on this timestamp. The model supports 40 symbols: 10 digits, 26 English alphabet letters and 4 special symbols(start of sequence symbol, end of sequence symbol, pad symbol and unknown symbol).

Note: that start and end symbols are not passed in the supported symbol set in the demo, as well as pad symbol. See parameter -m_tr_ss in the demo section for details.

Use text-detection demo

Model is supported by text-detection c++ demo. In order to use this model in the demo, user should pass the following options:

-tr_pt_first
-m_tr_ss "?0123456789abcdefghijklmnopqrstuvwxyz"
-tr_o_blb_nm "decoder_output"
-tr_composite
-dt simple -lower

For more information, please, see documentation of the demo.

Demo usage

The model can be used in the following demos provided by the Open Model Zoo to show its capabilities: