How to train a new model

[ImDePanDa]

Hi , Belval ! First , I want to say thanks.Your codes give me lots of help . But there are some problems with me . I use your pretrained model to recognise some nubmers and some symbols like"+" , but the results is not good，so I want to train my own model to do my work but I don't know how to train ,can you give me some guidance ? Hope for your reply.Thanks !!

[Belval]

Hi!

To retrain the model, it should be as simple as:

python3 run.py -m save/ -ex [TRAINING/TESTING DATA] --train

You need Tensorflow installed.

You can see python3 run.py -h for more tuning options.

[ImDePanDa]

@Belval Hi,belval! I have a new trouble.I use my own pictures to train a new model(hand-writen model),but I find it doesn't work.Here are some pictures about my data and wrongs.And I find there are somethings different about my data and your data.The second is your data's sequences and the third is my data's sequences.Can you give me some advices?

[ImDePanDa]

@Belval Hi,Belval! I have solved the above problem.The reason of this error is the names of the pictures are at the same length.But there is a new question.How can I use the model which has been well trained to predict a single picture?Look forward to your reply！

[Belval]

It used to be python3 run.py -ex [SAMPLE_FOLDER] --restore --test but I just tried it on my said an it did not work. Possibly something changed in TensorFlow/Numpy since then.

I'll try with older versions and see if I can get it to work.

[sushanth-d]

Hey! Thank you for all the code and all the explanations! :)

Any update on how to test single images? Or a set of images in a folder?

[Belval]

Haven't touched that repo in a while, currently on vacation. I'll add it to my todo list.