Official GitHub page of the journal publication "WEAR: An Outdoor Sports Dataset for Wearable and Egocentric Activity Recognition" to be published Issue 4 of IMWUT Vol. 8 (November 2024).
Hi! I've noticed here that you didn't batch at data loader level, but I've also read that it is the model itself in charge to do that. This means that each sample from your dataloader (which is the entire number of samples in a video) is fed into the model is a tensor with all the sequences from a single video, and then it's not possible to shuffle them. Am I missing something?
Hi! I've noticed here that you didn't batch at data loader level, but I've also read that it is the model itself in charge to do that. This means that each sample from your dataloader (which is the entire number of samples in a video) is fed into the model is a tensor with all the sequences from a single video, and then it's not possible to shuffle them. Am I missing something?