Can a single GPU complete the training of visdrone?

[kourlephy]

I only have a 3070 on my computer. When I started training, it kept showing "CUDA out of Memory". I would like to ask whether a single GPU can complete the training, thanks

[kourlephy]

I changed IMS_PER_BATCH, but I can't find the place to modify NUM_WORKERS. I guess that too large NUM_WORKERS may lead to insufficient CUDA memory. Could you help me to solve this problem?

[eiokobe]

hello, I also prepare to train on a single GPU，But now I have a problem, "SyncBN" needs to be changed to "BN", how do you change it? thank you!

[kourlephy]

hello, I also prepare to train on a single GPU，But now I have a problem, "SyncBN" needs to be changed to "BN", how do you change it? thank you!

I'm sorry that I haven't encountered your problem. At present, I can only wait for the author's reply to see what his explanation is:-)

[Yian-hao]

I only have a 3070 on my computer. When I started training, it kept showing "CUDA out of Memory". I would like to ask whether a single GPU can complete the training, thanks 你现在跑通了没？ 是在lunix系统上运行的是么，后来我使用了多卡，内存够了，但是会伴随有其他问题 最近作者更新了，还有在尝试么

[kourlephy]

I only have a 3070 on my computer. When I started training, it kept showing "CUDA out of Memory". I would like to ask whether a single GPU can complete the training, thanks 你现在跑通了没？ 是在lunix系统上运行的是么，后来我使用了多卡，内存够了，但是会伴随有其他问题 最近作者更新了，还有在尝试么

我没跑通，找了其他论项目改了写论文了，不过我们实验室有个师弟最近在跑这个，我帮你看看他能不能跑通吧

[Milkyway-xX]

I only have a 3070 on my computer. When I started training, it kept showing "CUDA out of Memory". I would like to ask whether a single GPU can complete the training, thanks 你现在跑通了没？ 是在lunix系统上运行的是么，后来我使用了多卡，内存够了，但是会伴随有其他问题 最近作者更新了，还有在尝试么

我在deepin和win11下都跑了，但是infer还在调试。win下可以去搜detectron2的安装方式。 可以参考的链接：https://zhuanlan.zhihu.com/p/584444690 希望能帮到你

[Milkyway-xX]

I only have a 3070 on my computer. When I started training, it kept showing "CUDA out of Memory". I would like to ask whether a single GPU can complete the training, thanks

3070可以试着调小yaml中的img per batch，我双3060可以在默认batch下跑通，单3060在batch=2的情况下看显存占用才7G左右，应该是能行的

[Yian-hao]

谢谢兄弟，作者更新完代码之后可以跑了，后来我又尝试了一下，训练和推断都行

[Yian-hao]

I only have a 3070 on my computer. When I started training, it kept showing "CUDA out of Memory". I would like to ask whether a single GPU can complete the training, thanks

3070可以试着调小yaml中的img per batch，我双3060可以在默认batch下跑通，单3060在batch=2的情况下看显存占用才7G左右，应该是能行的

I only have a 3070 on my computer. When I started training, it kept showing "CUDA out of Memory". I would like to ask whether a single GPU can complete the training, thanks 你现在跑通了没？ 是在lunix系统上运行的是么，后来我使用了多卡，内存够了，但是会伴随有其他问题 最近作者更新了，还有在尝试么

我在deepin和win11下都跑了，但是infer还在调试。win下可以去搜detectron2的安装方式。 可以参考的链接：https://zhuanlan.zhihu.com/p/584444690 希望能帮到你

谢谢，我在lunix上也跑通了。你调试infer调的怎么样，我发现运行完infer并不会形成识别后的效果图，这个挺让人苦恼的

[Milkyway-xX]

I only have a 3070 on my computer. When I started training, it kept showing "CUDA out of Memory". I would like to ask whether a single GPU can complete the training, thanks

3070可以试着调小yaml中的img per batch，我双3060可以在默认batch下跑通，单3060在batch=2的情况下看显存占用才7G左右，应该是能行的

I only have a 3070 on my computer. When I started training, it kept showing "CUDA out of Memory". I would like to ask whether a single GPU can complete the training, thanks 你现在跑通了没？ 是在lunix系统上运行的是么，后来我使用了多卡，内存够了，但是会伴随有其他问题 最近作者更新了，还有在尝试么

我在deepin和win11下都跑了，但是infer还在调试。win下可以去搜detectron2的安装方式。 可以参考的链接：https://zhuanlan.zhihu.com/p/584444690 希望能帮到你

谢谢，我在lunix上也跑通了。你调试infer调的怎么样，我发现运行完infer并不会形成识别后的效果图，这个挺让人苦恼的

整了一天还没通。。卡在train_tools的coco_infer.py，有个import detectron2_backbone 没这个库，还不知道从哪来的。请问您这边infer能通吗

[huangjiping]

I only have a 3070 on my computer. When I started training, it kept showing "CUDA out of Memory". I would like to ask whether a single GPU can complete the training, thanks

3070可以试着调小yaml中的img per batch，我双3060可以在默认batch下跑通，单3060在batch=2的情况下看显存占用才7G左右，应该是能行的

谢谢兄弟，作者更新完代码之后可以跑了，后来我又尝试了一下，训练和推断都行

你好，我在运行python visdrone/data_prepare.py --visdrone-root data/visdrone时出现了No such file or directory: 'data/visdrone/coco_format'这个问题，我看官方也没有coco_format.py这个文件，请问你是如何解决这个问题的啊？

[LaplaceSama]

That's strange. I tied it out on rtx3070 too. But it kept telling me the version of my cuda was too high(The required CUDA version is 10 but mine is 12). I tired to low the version of CUDA but version was still incompetent. It was until I turned to use an older GPU(single) success. I wonder how your environment setting is. 奇了怪。我也在3070上做过配置，但总报错说我cuda版本太高。哪怕我降低cuda版本后还是报错。后来我用了旧的GPU才配置好。想请问一下你是怎么做的配置？

[ahfevounccfvsa]

电脑上没有gpu，想利用kaggle上的gpu跑可以吗，但是在kaggle上一直没有办法创建新的环境 (我不太会创建) conda querydet 一直报错解决不了，有没有懂的uu