Dataset and Results

[songeza]

1.The original RUWI training set and test/validation set consisted of 400 and 300 images, respectively. Could you please explain why in this paper the training and test sets consist of 525 and 175 images, respectively? Could you please provide more details about the dataset, such as the specific format of the train.txt and val.txt files?
2.To enhance the impact of the paper and avoid reinventing the wheel, it has been customary for your team (led by Prof. Huchuan Lu) to share results of comparative methods in previous works. Can you confirm if the authors have plans to do the same in this paper?

[Drchip61]

First and foremost, I apologize for the delayed response. Due to some personal family issues and my habit of procrastination, I have not been able to reply to your queries promptly. I am sincerely sorry for any inconvenience this may have caused and will address your questions as soon as possible.

Regarding your inquiry about the RUWI dataset splitting, here are the details:

For the MAS3K, RMAS, and UFO120 datasets, we followed the default splits as these datasets already have established methods and results. However, for the RUWI dataset, which lacks existing methods, we used an approximate 5:1 split. During the initial submission, we were under time constraints, and after using the 5:1 split, we did not change it later.

In addition, concerning the visual results you mentioned, we are currently organizing the data and will release the results soon. Again, I apologize for the delay.

I hope this explanation clarifies your concerns. If you have any further questions or need additional clarification, please do not hesitate to let me know.

[Drchip61]

Regarding your inquiry about the dataset splitting, I will upload the details within 3 hours of this email.