Use Case and High-Level Description

This is a network for handwritten Japanese text recognition scenario. It consists of a VGG16-like backbone, reshape layer and a fully connected layer. The network is able to recognize Japanese text consisting of characters in the Kondate and Nakayosi datasets.

Example

-> 菊池朋子

Specification

Metric	Value
GFlops	117.136
MParams	15.31
Accuracy on Kondate test set and test set generated from Nakayosi	98.16%
Source framework	PyTorch*

Accuracy Values

This demo adopts label error rate as the metric for accuracy.

Inputs

Shape: [1x1x96x2000] - An input image in the format [BxCxHxW], where:

B - batch size
C - number of channels
H - image height
W - image width

Note that the source image should be converted to grayscale, resized to specific height (such as 96) while keeping aspect ratio, and right-bottom padded.

Outputs

The net outputs a blob with the shape [186, 1, 4442] in the format [WxBxL], where:

W - output sequence length
B - batch size
L - confidence distribution across the supported symbols in Kondate and Nakayosi.

The network output can be decoded by CTC Greedy Decoder.

Legal Information

[*] Other names and brands may be claimed as the property of others.