Open duduheihei opened 3 years ago
I use TimesFormer to train on my own dataset, the input channel is 1, num_frame is 8. I got nearly 46% accuracy, I guess it was because imbalance of the dataset classes. I want to know if anyone has trained this model sucessfully?
I am waiting for the pre trained weights from the official repo to get started. Eager to see how well it will perform compared to slowfast ...
I use TimesFormer to train on my own dataset, the input channel is 1, num_frame is 8. I got nearly 46% accuracy, I guess it was because imbalance of the dataset classes. I want to know if anyone has trained this model sucessfully?