Skip to content

Latest commit

 

History

History
40 lines (25 loc) · 1.48 KB

README.md

File metadata and controls

40 lines (25 loc) · 1.48 KB
Raw

text-detection-0003

Use Case and High-Level Description

Text detector based on PixelLink architecture with MobileNetV2-like as a backbone for indoor/outdoor scenes.

Example

Specification

Metric Value
F-measure (Harmonic mean of precision and recall on ICDAR2015) 82.12%
GFlops 51.256
MParams 6.747
Source framework TensorFlow*

Inputs

Image, name: Placeholder, shape: 1, 768, 1280, 3 in the format B, H, W, C, where:

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

Expected color order: BGR.

Outputs

  1. name: model/link_logits_/add, shape: 1, 192, 320, 16 - logits related to linkage between pixels and their neighbors.

  2. name: model/segm_logits/add, shape: 1, 192, 320, 2 - logits related to text/no-text classification for each pixel.

Refer to PixelLink and demos for details.

Legal Information

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