I was trying to train for this BN-WVAD model as the author trained for 2 datasets, the author had only released the codes for 1 of the dataset (XD-Violence).
I had successfully trained for this dataset, but failed to train for the 2nd dataset (UCF-Crime), keep getting the CUDA out of memory problem.
I had tried using my own RTX3050 of 6GB, Google Colab’s T4 GPU and Lightning AI Studio’s A10G which has 24GB VRAM but still getting same issue as shown in the pictures below.
Previously had checked the shape of both dataset (N, 1024). Seems like xd-violence is greater than ucf crime
I wonder why I can successfully trained for xd-violence when the shape is greater. Any helps will be much appreciated