I've created a dataset in my master's thesis (Thesis Evaluation of the Performance of Deep Learning Techniques Ov...
) that I called it puzzle dataset from natural images with 7 categories. Each Category has 36 to 40 images and that's a small dataset to be used in deep learning methods. For this reason, we came up with a new idea which was dividing one image into 100 pieces, so each category would have between 3600 images to 4000 images. That amount of data would be enough to train your model.Therefore, do you think that this was a good way of creating a dataset or there's a better way to create it?. Plus, do I have to publish it, so it will be officially mine.
Looking forward to seeing your suggestions and ideas regarding this topic!
Are you sure that you will save the quality of the dataset by dividing each image into 100 parts? This way you can increase the noise level of input data. I would recommend searching for "similar" images, or generating synthetic images based on the existing dataset.
Dear Imran Jamal
;Thanks a lot for your reply. I really appreciate your advice and I'll follow it in order to publish my dataset too.
Dear Igor Gajniyarov , I already published my thesis with my puzzle dataset. I will consider what you've said for future works. Thanks for your reply.
Dear Ihab A. Satam , I'll work on it immediately. Thanks a lot for your reply.
what is your consideration when you dividing the dataset? and how you do that?
Titan Muslim Al-Khan I just mentioned that in my question above!
Titan Muslim Al-Khan Thank you for contributing to my question. I really appreciate that.
Off course, you can contribute the dataset alongside the published article or even as an independent dataset if your dataset is generic in nature and can be applied in similar scenarios. However, there are always issues of simulated datasets which sometimes reflects the individual control over the direction of the data and therefore, the journals' reviewers take a very keen look on the applicability of the dataset.
Shabir Ahmad Thanks a lot for your reply!. I will make sure that my dataset is generic in nature and can be applied in similar scenarios.
- Badges
- Science topic
- Similar topics
- Computer Science
- Images