ankanbhunia / handwriting-transformers Goto Github PK

How to get the test set

Hi,

Thanks for sharing the great work!

I am training this model from scratch, but the training parameters mentioned in the paper, such as IGM_HEIGHT and LR are different than the ones given in the params.py file.

Can you please share the parameters that were used to train the network?
Thanks.

Hello!

I need to generate color images for my project. I want to try changing your architecture accordingly. I guess this should be easy to do. It is enough to change the number of channels at input and output, right? I'm wondering what you think of this possibility. Any thoughts welcome!

Great project. I saw the paper had experiments with RIMES, any chance you still have those weights? Thanks!

Sir, the method proposed in your paper has effectively improved.
I am very curious how to get the "fid: four scenes" in your paper
In addition, Would you like to share the pre-trained model.
I am sorry for my recklessness.
good luck.

Hello, I have been troubled by some questions for a long time, and I hope to get answers. Regarding the parameters in prapare_data.py: TRAIN_IDX = 'gan.iam.tr_va.gt.filter27' and TEST_IDX = 'gan.iam.test.gt.filter27', how were they obtained? Are the IAM_WORD_DATASET_PATH and XMLS_PATH directories used to store word images and related XML files, respectively, downloaded from the I AM dataset official website?
I hope to receive your answer, and thank you very much.

每次在训练后加载权重文件时候报错RuntimeError: Error(s) in loading state_dict for Generator:
Missing key(s) in state_dict: "encoder.layers.0.self_attn.in_proj_weight"

Hello, thanks for the impressive work.
It seems some code(lexicon file and txt not be uploaded) and the install.md not be finished.
When will this code be updated?
@ankanbhunia

@ankanbhunia

While you mention:
"You can train the model in any custom dataset other than IAM and CVL. The process involves creating a dataset_name.pickle file and placing it inside files folder. The structure of dataset_name.pickle is a simple python dictionary."

I go through the codes and the way you suggest seems to retrain the model on another dataset from scratch rather than fine-tuning your model. As your paper has not yet much discussed this, I come to ask your opinion. If I want to apply the model to generate other language's handwritings e.g. Japanese, is there a way to quickly fine-tune your model upon a new Japanese handwriting data? Or do I need to retrain it from scratch?

Hi, sir, thanks for the impressive work. Mentioned in the paper, "To retain information regarding the order of input sequences being supplied, we add the positional encodings [23] to the input of each attention layer". However, the released code does not add the positional encodings to the Multi-Head Attention of the encoder, and only adds positional encodings to the Multi-Head Attention of the decoder. It's better if we don't apply positional encodings in the encoder?

Kindly please give the steps in detail, in regards to how one can use their own handwriting as data to generate the same handwriting, many people are interested in this but due to complexity unable to use this beautiful model

Hello author, thank you for your excellent paper. I have recently looked at your code and I have encountered a problem that I am not generating Fake and Real images in the saved_images folder after the training is completed, these two folders are empty, I would like to get your guidance and help, looking forward to your reply, thank you!

Hi, many thanks for your innovative work, just wondering is there a plan for code release? thanks

Thank you for publishing IAM_32.pickle in Issue #4
I cloned this project to my personal Colab Notebook and executed "train.py". Now the file CVL_32.pickle is required.

Traceback (most recent call last):
  File "./Handwriting-Transformers/train.py", line 42, in <module>
    TextDatasetObjval = TextDatasetval()
  File "/content/Handwriting-Transformers/data/dataset.py", line 109, in __init__
    file_to_store = open(base_path, "rb")
FileNotFoundError: [Errno 2] No such file or directory: '../CVL_32.pickle'

Would you please publish this file also?

Thank you very much. 🙇

Hello, what's in your code ‘base_path = '../IAM_32.pickle'’ 。Can this document be published? If so, please send me a private letter? Thanks for your reading！

Love this work! Any chance you could add an MIT license or something? I've started my own fork, but wanted to make sure it remains open so people can keep building off of it. Thanks!

Hello, what's in your code ‘base_path = './CVL_32.pickle'’ 。Can this document be published? If so, please send me a private letter? Thanks for your reading

Hi!

I've been playing around with this model locally following the instructions in the README and my results don't seem to be nearly as good as yours.. I'm following your instructions here #11 (comment) and then running the prepare.py in my fork

For instance even with different style prompts the model seems to generate very similar results for me
Real on left, Generated on right

IAM style 1

IAM style 2

and then secondly the CVL and the IAM models give very different results to each other. But also quite consistent results within the model itself for different styles

CVL style 1

CVL style2

Is there something stupid I'm missing or do I need to train it with these writers in the dataset to get better results? does the google drive contain the fully trained models that were used to generate the results in the paper?

Very cool project though - congrats!!

I went through the codebase, it seems like the cycleloss has been commented. is there any particluar reason why it has been commented out?

I'm trying to generate something based on my own handwriting, what text cropping tool did you use? Wondering if I can skip the work of cropping every single word by hand.

Thanks!!

Could this be used to reproduce one particular handwriting style? (That is, one not in the IAM database -- such as my own handwriting, or the handwriting of a famous historical figure.)

If so, could you please walk me through how to do that? I would be very appreciative.

warnings. warn(
C:\Users\Asus\anaconda3\envs\pytorch\lib\site-packages\torchvision\models_utils.py:223: UserWarning: Arguments other than a weight enum or None for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing weights=ResNet18_Weights.IMAGENET1K_V1. You can also use weights=ResNet18_Weights.DEFAULT to get the most up-to-date weights.
warnings.warn(msg)