hello！I have some questions about run360.py

[wrryyy1]

Dear author, thank you for your contribution and sharing. 😚 Currently, I have conducted an experiment to reproduce the data in Table 1 of the paper. I used NVIDIA A100 (with 80G of VRAM) to run run360.py, and the dataset was Mipnerf360. I found that the scenes (bicycle, garden, stump) with the MV_largest=[1,1,1,3] setting could be trained normally, but the scenes (such as room, kitchen, etc.) with the MV_normal = [48, 24, 12, 8] setting would report an out-of-memory error during the pre-training stage. 😵 Why does Run360 actually require over 70GB of VRAM? Could you please explain why this happens and if there are any follow-up settings or modifications? 😴 How should I change the training settings? Thank you! 😢