custom_dataset

[scream68]

I'm trying to train using a custom dataset, and sometimes the number of actual labels in the scene is less than 10 (the predicted count). Should I perform padding or any other processing? If so, how should I do it? For example, if I'm predicting two classes, 0 and 1, and I set num_classes=2, before calculating the loss, the code will fill it as [1, 1, 2, 0, 2, 2, 2, 0, 1, 2]. such as the figure

[lio5474]

Hello, have you succeeded? I also want to use a custom dataset to run maptr. How can I build my own dataset, which means what data and annotation information are needed?