显存问题

[xiaohongzaizhe]

我使用了48G的显存的显卡，parser.add_argument('-chunk', type=int, default=5 , help='crop volume depth') #原本96
parser.add_argument('-num_sample', type=int, default=1 , help='sample pos and neg')
parser.add_argument('-roi_size', type=int, default=96 , help='resolution of roi') parser.add_argument('-b', type=int, default=1, help='batch size for dataloader')都会爆显存，而太低的参数会导致loss=1 模式崩溃

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

Title: Video memory problem

[xiaohongzaizhe]

请问原作者是在单卡显存多少的情况下训练的？

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

May I ask how the original author trained with the amount of video memory on a single card?

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

How can I support Doka training?

[WuJunde]

try smaller batch size or image resolution

[2308700388]

Can 3090 be trained