Spaces:
Runtime error
Availability of pretrained model
Is the pretrained model used in the hugging face space demo is available to download.
Hi, the space uses three object detectors for coarse layout analysis, cell detection and row/column detection. Cell and row/column detection is available for download as it is. Layout analysis detector is available for download in a different version trained on a less diversified dataset for a shorter time. Check https://github.com/deepdoctection/deepdoctection which sets the foundation to this repo. You can also fine tune the layout model using your custom dataset in order get more accurate results for the domain you are interested in.
@JaMe76
The layout results in this space are pretty interesting and far better than using the default layout weights (layout/d2_model_0829999_layout_inf_only.pt). I know we can't have the model for privacy reasons, but could you please tell me which model architecture you used, so that I can further train it myself? Is the same one in d2_casc_rcnn_X_32xd4_50_FPN_GN_2FC using detectron2?
@JaMe76 , why don't you release the model which are running on Hugging Space?
@anthony2261 I think he has mentioned in one of his github issues that the model is same, however the dataset used is different (PubLayNet + DocLayNet + 3k private images).
@yashsandansing , I am successfully able to crack his Hugging Face Code. Thanks @JaMe76 . Its a great learning.
@shaileshsarda can you share your solution ?