Akarsh comments

Results 105 comments of


                                            Akarsh

Using pre-trained models

I am looking at the approach in this way, if the model can overfit that means, the implementation is going in the right direction. Now, the idea of pre-training came,...

Using pre-trained models

Although, I didn't get, when you say "there is no attempt to save the embedding features or visual features"

Using pre-trained models

Not till now (since busy with one of my research work)

Using pre-trained models

Hi @riteshKumarUMass @HoomanKhosravi, can you guys again try the fine-tuning using the pre-trained weights (I have attached them in the readme)

We can not find DocFormer_For_IR in modeling.py and modeling_l.py

Hi @yuni1314, actually when implementing DocFormer for IR, the author didn't exactly mention the decoder, so i didn't go forward with it.

will you release any pretrained model?

We are trying to think for releasing it, but we need to make sure that, our model is performing as closely as that mentioned in the paper

will you release any pretrained model?

Hi @philschmid, I did try to train DocFormer from scratch on FUNSD dataset. Here is the script, [Kaggle Notebook](https://www.kaggle.com/akarshu121/docformer-for-token-classification-on-funsd), and here are the results [Weight and Bias Report](https://wandb.ai/iakarshu/DocFormer%20on%20FUNSD?workspace=user-iakarshu) When we...

Downloading the Image for the whole dataset

Hi @furkanbiten, can you tell me the steps/ refer to appropriate location to download the pdfs. Here is what I am able to see. ![image](https://user-images.githubusercontent.com/55104596/206382633-5f621fdb-537a-4cb9-81fe-787d458f0f29.png) Regards, Akarsh

Downloading the Image for the whole dataset

Got it. I would try and let you know. Thanks, Akarsh