Custom dataset

[Eurekaer]

I apologize for bothering you again. Recently, I have been working on customizing a dataset similar to UCF101-24. Do you have any good suggestions, such as annotation tools, etc.? During my search, I came across a similar tool link. Additionally, if I want to train on a custom dataset similar to UCF101-24, are there any specific things I should pay attention to? Generally speaking, when training on a custom dataset, which dataset format is better between UCF101-24 and AVA?

[Hope1337]

@Eurekaer

• About the annotation tool: I have never used any tool for annotation, but there was one time when I came across this tool; you may consider trying it.

• Regarding some notes for training on a custom dataset:

  • Ensure that you use index labels starting from 0.
  • Adhere to the format (e.g., tensor shape) outlined in my guide note.
  • Remember to update num_classes in the config file, along with data_root, idx2name and dataset.

• I think UCF101-24 format will be more convenient

If you encounter any issues during the training process or if the model produces unexpected or poor results, please don't hesitate to provide feedback to me.

[Eurekaer]

Thank you very much for your reply. I wish you a happy life!