j-rausch / dsg Goto Github PK

Is there any setting that can be added during the execution of demo.py to include the div for the graphic figures contained within the image in the hOCR, or is it automatic? Because I'm testing it on various images, and currently, it's not being detected. However, I noticed that it's present in sysdemo.

Half the repos in the installation process are not found, pytorch 1.11 is no longer served by pip, neither is the numpy version.

Is there a simpler installation process to your knowledge? I would really like to try your code but the wall of installs is preventing me from going forward.

Thank you for publishing this great work.
I successfully run a demo using the Eperiodica dataset.
But, I couldn’t find the arXiv dataset in your Google Drive link.
Is it possible to publish the arXiv dataset?
Thank you in advance.

Hi, Thank you for the great work.

I see that the repo only has Prediction/demo codes available. Can you please add Finetuning code and Custom Dataset format that the model accepts. Can you also please add all details required to custom train on the datasets.

Thank you

Hi, how can I get the text of a image through OCR? Is there already something implemented, or do I need to do it outside of this project?

Hi, and thank you for this great project.

I am trying to test on my own image, and I don't know how to generate the accompanied .txt file for the image.
I see there is one in the sysdemo folder.
How do I generate this file?
Thank you.

Thanks for your great project! I would like to utilize the datasets you've provided for my research. However, I'm encountering some difficulty in interpreting the rules for relationship annotation within the provide files. Could you please clarify the annotation guidelines for me?

Why can a figure have so many parents? Also why can a row be the parent of so many entities?

Hi, thank you for this great project!
I'm trying to test my own images and I got OCR file using Tesseract, but the format is different from the OCR file in sysdemo.
What do the two numbers at the beginning of the OCR file in sysdemo mean? And are the four numbers following each word the coordinates?
I look forward to your reply.

Hi, Johannes, thanks for your great project! I want to follow your datasets and work to do research. However, I haven't found the evaluation scripts to reproduce the results in DSG paper.
I tried to modify the training scripts like:
CUDA_VISIBLE_DEVICES=0 python scripts/train_doc_SG_head.py \ --config-file ./configs/sgg_end2end_EP.yaml \ --num-gpus 1 \ --eval-only \ --resume \ MODEL.ROI_SCENEGRAPH_HEAD.PREDICT_USE_VISION True \ OUTPUT_DIR ./output/eval \ MODEL.WEIGHTS ./checkpoints/DSG_E2E_eperiodica/dsg_e2e_eperiodica_checkpoint.pth
However, it reported a bug:
TypeError: _evaluate_predictions_on_coco() got an unexpected keyword argument 'use_fast_impl'
I do not know what is the correct command.

Thanks! Look forward to your reply!