How to train my images dataset with crnn_vgg16_bn

[Shivashankarar]

I wwant to train my image dataset. How can I train it?
Can you please explain the steps , @felixT2K ?

[felixT2K]

Hi @Shivashankarar 👋🏼 ,

Please do not open further discussions with the same question. I have already replied to this in #1247 (reply in thread) :)

Please take a look at the data format section: https://github.com/mindee/doctr/tree/main/references/recognition#data-format
Each image should contain only 1 word for example: https://www.mdpi.com/sensors/sensors-22-06025/article_deploy/html/images/sensors-22-06025-g016.png

Let's say the image with 0.0019 is img_1.jpg in a images folder than the corresponding labels.json would look like:

{
    "img_1.jpg": "I",
}

Overall structure would be:
train_dataset/
images <-- contains all images
labels.json <-- contains all labels

(same for the validation dataset)

After your dataset is prepared you can run:
python3 references/recognition/train_pytorch.py crnn_vgg16_bn --train_path=<PATH_TO_TRAIN_SET> --val_path=<PATH_TO_VAL_SET> --pretrained