Request for Release of Bounding Box Coordinates

[KelvinBaconBrown]

Dear Authors,

Thanks for your great work! I was wondering if it would be possible for you to release the bounding box coordinates for each image used in MedTrinity-25M. Access to this information would greatly facilitate further analysis and enable additional research opportunities.

[sahil02235]

Dear Authors,

I would like to respectfully suggest that you consider re-releasing the dataset with the original images and their corresponding bounding box coordinates. The current practice of overwriting images with bounding boxes considerably restricts the dataset's utility for a broader range of research applications.

By providing the original images alongside the bounding boxes, you would greatly enhance the dataset's versatility. This approach would enable researchers to employ the data in various innovative ways, such as generating bounding boxes and corresponding captions from the original images. Such flexibility is crucial for advancing research in areas like image captioning and object detection, where generating or manipulating annotations can lead to valuable insights and improvements in model performance.

Incorporating original images allows for additional research opportunities, such as exploring new methods for automatic annotation and evaluating the effectiveness of different bounding box generation algorithms. It also facilitates comparative studies where different techniques can be tested on the same set of images and annotations.

Thank you for considering this suggestion, which I believe would contribute significantly to the ongoing advancement of research in the field.

[yunfeixie233]

Dear @KevinBaconBrown and @sahil02235,

I sincerely apologize for any inconvenience caused. Due to a current deadline, I am unable to address these issues immediately but will give them my full attention next week.

Thank you

[sahilqure]

@yunfeixie233 Thanks for your consideration.
Let me know if I can help you in any way regarding this.
Email: [email protected]

[sahil02235]

@yunfeixie233 Lot's of medical conferences deadline are coming and I wanted to use this data. Can you please help us with the clean image and bounding boxes?

[yunfeixie233]

Dear @sahil02235 and @sahilqure ,

I apologize for not responding sooner; I was in Milan attending ECCV over the past few days. I’m now back and fully focused on the tasks at hand. I'll begin checking the data immediately and plan to provide updates by Wednesday, if everything is ready.

Thank you!

[yunfeixie233]

Dear @sahil02235 and @sahilqure,

Thank you for your patience.

We didn’t originally plan to release the bounding boxes, and part of the raw files were lost due to a server hack. While recovery is ongoing, you can use this tool to extract more than 90% of the bounding box coordinates from metadata: [https://huggingface.co/datasets/UCSC-VLAA/MedTrinity-25M/blob/main/toolkit/bbox.py].

Once the server is restored, I will provide the original files. Apologies for the inconvenience, and thank you for your understanding.

Thank you!

[Ben81828]

Dear @yunfeixie233,

Thank you very much for your work.

Will the original files that you release later include raw images without any bounding box annotations?
I believe this would be very helpful for training a model that takes raw images as input and outputs the corresponding bounding box information.