cuda out of memory for a RTX 4090

[Candy-Crusher]

Hi! I'm trying to reproduce this work. As said in the paper, you used a Quadro RTX 4000 laptop GPU, whose memory size is 8GB. For DSEC-Det dataset, you used bs=8 for inference and bs=32 for training. But when I ran the code, I found that the memory usage for inference with bs=1 is about 13-20GB. Is there any parameter that I set wrong?