https://github.com/shambac/project-sem-6

Handwritten text recognition with LSTM RNN using IAM dataset
https://github.com/shambac/project-sem-6

handwritten-text-recognition tensorflow

Last synced: 2 months ago
JSON representation

Handwritten text recognition with LSTM RNN using IAM dataset

• Host: GitHub
• URL: https://github.com/shambac/project-sem-6
• Owner: ShambaC
• Created: 2023-03-11T05:54:59.000Z (over 3 years ago)
• Default Branch: main
• Last Pushed: 2023-03-13T17:44:16.000Z (over 3 years ago)
• Last Synced: 2024-12-24T11:52:20.883Z (over 1 year ago)
• Topics: handwritten-text-recognition, tensorflow
• Language: Python
• Homepage:
• Size: 17.6 KB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Project SEM 6

Handwritten text recognition

## What ?

Handwritten text recognition using Long Short Term Memory implementation of RNN. Honestly I don't know what's happening here.

The [IAM Handwriting Database](https://fki.tic.heia-fr.ch/databases/iam-handwriting-database) is used to train the network. Specifically the words dataset.

## How to ?

Clone the repo and then do the following.

### Get the IAM Handwriting Dataset :

- Make an [account](https://fki.tic.heia-fr.ch/register) on the database website.

- Download the [words archive](https://fki.tic.heia-fr.ch/DBs/iamDB/data/words.tgz)

- Extract the contents to the following directory structure on the root of this project : `Datasets/words`

    - So the resulting folder structure should be : 

        ```

        Datasets

            |_words

                |_a01

                    |_a01-000u

                        |_images.png

                    |_a01-000x

                    |_...

                |_a02

                |_...

        ```

- Now you need the labels for these images.

- Download the [ascii archive](https://fki.tic.heia-fr.ch/DBs/iamDB/data/ascii.tgz)

- Extract only the words.txt file to the `Datasets` folder.

### Install the dependencies

- install the requirements.

- Do in terminal : `pip install -r requirements.txt`

- The required packages are :

    - numpy : 1.22.4

    - opencv_python : 4.6.0.66

    - pandas : 1.3.2

    - Pillow : 9.4.0

    - tensorflow : 2.10.0

    - tqdm : 4.62.3

### Edit the configurations

- Open the `config.py` file.

- Make whatever changes you want.

- Mainly you might want to make changes to the following variables : 

    - batch_size

    - learning_rate

    - train_epochs

    - train_workers

    - validation_split

### Train the model

- Run the `train.py` script to train the model.

- No need for editing anything in this file.

- I trained the model with the following parameters :

    - learning rate : 0.0005

    - validation split : 0.9

    - batch size = 16

    - epochs = 100

- I trained it with my GTX 1650. It used 2132 MB of GPU memory. Usage was around 8-10 %. Took around 140-150 seconds each epoch. Took about 4 hours for all 100 epochs.

- CPU usage was around 50%. My CPU is Ryzen 7 3750H.

- RAM usage was around 2 GBs.

- You can visualise the training using tensorboard. Run `tensorboard --logdir = path_to_logs` in terminal to start the server.

- The logs are located at the following folder : `Models/Handwriting_recognition/{timestamp}/logs`

- After training, a bunch of files are generated. The only important files are `model.meow` and `configs.meow`.

### Run the model

- You can run the `inferenceModel.py` to check with only one image from the dataset itself.

- OR use the Paint GUI made with tkinter by running the `tkRecog.py`.

# IN BOTH CASES MAKE SURE TO EDIT THE unixTime VARIABLE WITH YOUR MODEL'S FOLDER

# Thank You

Totally not copied code.