Issue detecting dense object

[ladanisavan]

Hi there

I am using newly added regnetx_4gf_dds_FPN from "new_baselines" to detect words from A4 size documents. The issue that I am facing is after training on a good amount of data (300k samples) model is not able to detect all the words from the document image. I have tried multiple things but nothing worked. Any help would be appreciated.

Architecture/model used (without any change in configuration) __regnetx_4gf_dds_FPN__

Things I have tried

• increasing limit of __model.roi_heads.box_predictor.test_topk_per_image__ from 100 to 2000 (which would be a sufficient IMO)
• increasing the __model.proposal_generator.pre_nms_topk__ to (2000, 8000)
• increasing the __model.proposal_generator.post_nms_topk__ to (1000, 8000)
• model input image size is changed to 1024 x 1024
• trained for 50 epochs

Here is a sample prediction

[github-actions[bot]]

You've chosen to report an unexpected problem or bug. Unless you already know the root cause of it, please include details about it by filling the issue template. The following information is missing: "Instructions To Reproduce the Issue and Full Logs"; "Your Environment";

[ppwwyyxx]

As the issue template mentions:

If you expect a model to converge / work better, we do not help with such issues, unless a model fails to reproduce the results in detectron2 model zoo, or proves existence of bugs.

We provide configs & models with standard academic settings and expect users to have the knowledge to choose or design appropriate models & parameters for their own tasks.