clovaai / deep-text-recognition-benchmark

Text recognition (optical character recognition) with deep learning methods, ICCV 2019
Apache License 2.0
3.77k stars 1.11k forks source link

Training TPS-ResNet-BiLSTM-Attn.pth model on my own custom dataset to detect text and numbers from pictures of shipping container ids #419

Open venomvas opened 6 months ago

venomvas commented 6 months ago

Good Afternoon,

As the title says I want to fine tune the model to detect with better accuracy for my images of shopping container ids. images that are like these: id_27 id_1

One picture for example is vertical and the other is horizontal.

My questons are below:

1) Should I preprocess my pictures? 2) How should I organize my database . For example should I have pictures like these for train, validation and, testing ? and how should I write it in the terminal when finetuning 3) Inside he text file, the labels should be like this? : id_27.jpg MSCU 629071 0 id_1.jpg M S C U 6 2 9 0 7 1 0 Because the first picture is horizontal and the letters have spacing should they be written as above? And the second picture is vertical, should I consider giving each letter/number a space? 4) Can I have an example on how to do the testing of my model after I fine tuned it?

Thank you in advance, Vasilis Vasileiou