I am currently trying to process event data using a fixed 50ms time interval because my label data is annotated at 50ms intervals. However, I have encountered an issue: if there are gaps of several frames in the label data—for example, there is label data at 1000ms, and the next label data appears at 1500ms—the event data generated between 1000ms and 1500ms will have no corresponding labels, which are known as negative samples. This causes problems during training because the dimensions of the label data do not match those of the event data. How can I solve this problem while maintaining a fixed time interval for the event data?
I am currently trying to process event data using a fixed 50ms time interval because my label data is annotated at 50ms intervals. However, I have encountered an issue: if there are gaps of several frames in the label data—for example, there is label data at 1000ms, and the next label data appears at 1500ms—the event data generated between 1000ms and 1500ms will have no corresponding labels, which are known as negative samples. This causes problems during training because the dimensions of the label data do not match those of the event data. How can I solve this problem while maintaining a fixed time interval for the event data?