This is an text-recognition composite model that recognizes scene text. The model uses predefined set of alphanumeric symbols (case-sensitive) to predict words. The model is built on the ResNeXt-101 backbone with additional 2d attention-based text recognition head.
|Accuracy on the alphanumeric subset of ICDAR13||0.8995|
|Accuracy on the alphanumeric subset of ICDAR03||0.9389|
|Accuracy on the alphanumeric subset of ICDAR15||0.7355|
|Accuracy on the alphanumeric subset of SVT||0.8764|
|Accuracy on the alphanumeric subset of IIIT5K||0.8413|
|Text location requirements||Tight aligned crop|
The above accuracies are calculated for case-insensitive mode (i.e. GT text and predicted text are all casted to lowercase).
The text-recognition-0015-encoder model is a ResNeXt-101 like backbone with convolutional encoder part of the text recognition.
1, 1, 64, 256 in the
1, C, H, W format, where:
C- number of channels
H- image height
W- image width
1, 1, 1024. Initial context state of the GRU cell.
1, 16, 1024. Features from encoder part of text recognition head.
The text-recognition-15-decoder model is a GRU based decoder with 2d attention module.
1. Previous predicted letter.
1, 16, 1024. Encoded features.
1, 1, 1024. Current state of the decoder.
1, 1, 1024. Current context state of the LSTM cell.
1, 66. Classification confidence scores in the [0, 1] range for every letter.
Model is supported by text-detection c++ demo. In order to use this model in the demo, user should pass the following options:
For more information, please, see documentation of the demo.
[*] Other names and brands may be claimed as the property of others.