Comments (3)
Hi @francotheengineer! Thank you for your great question.
The RNN steps directly correlate to vertical slices of the input image, which is a cropped word. The first few steps of the RNN contain character inferences based on very little information, as only a few slices of the input image are included. So we do not include them in inference. However, those steps are not exactly fully discarded. The output of step N affects the output of step N+1, and so on. So even though we do not care to use the character predictions from Step 1, the state of Step 1 was included in calculating Step 2, which was used in calculating Step 3, which we do use directly for inference. Removing the first couple steps is used in most examples for convolutional recurrent neural networks for OCR (e.g., see the decode_batch
function in the Keras OCR example).
Does that answer your question? I realize it lacks a robust theoretical underpinning (e.g., why discard two steps? Why not three?). But there's certainly no reason why you couldn't try using the model with rnn_steps_to_discard=0
to see for yourself how this affects the quality of the recognition step.
from keras-ocr.
Closing this issue as it is primarily a Q&A matter as opposed to a bug or feature request. If I'm mistaken, please do comment back and we can re-open. Thanks!
from keras-ocr.
Thanks so much for the detailed help! Thanks so much! Franco
from keras-ocr.
Related Issues (20)
- Loading trained model doesn't work in new session - shape mismatch error.
- Can load pretrain custom digits model
- Predicted result strings are empty --> Predicted: ['', '', '', '', '', ''] HOT 1
- how to add keras-ocr new model?
- Error when training detector with batch size > 1
- local variable "image" out of scope in keras-ocr/tools.py
- Keras_ocr library not detecting specific texts from the image
- Please Help i cant display the evaluation score after prediction
- How to predict numbers only? HOT 1
- Evaluation of my model HOT 2
- Python 3.10 support HOT 2
- What version of matplotlib are you guys using?
- Could not download HOT 1
- Very inaccurate results with keras-ocr tflite model HOT 1
- Cannot train Recognizer - ValueError: Exception encountered when calling layer 'lambda_1' (type Lambda). HOT 5
- "Tried to convert 'num' to a tensor and failed. Error: None values not supported." HOT 1
- Can I get Korean Text from Image? Using keras-ocr HOT 1
- Open Source License HOT 1
- Adding an example for fine-tuning both detector & recognizer using an your own dataset HOT 4
- Detecting vertical text with craft HOT 3
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from keras-ocr.