CamVid DATASET

[BUAA-LKG]

Excellent work! Could you provide a link to download for the 11 category camvid dataset?

[BUAA-LKG]

Because I was curious how such high accuracy could be achieved on this dataset

[XuJiacong]

You can directly download the full dataset via: https://www.kaggle.com/code/mahmudsarwar/image-segmentation-pytorch/data?select=CamVid ,then unzip it in data/camvid dir. Since the work was implemented by myself, I am also very surprised by the results, so if you can help me to double check it, I will be appreciated.

[XuJiacong]

Basically, you can check: 1) if the colors (rgb) for the 11 categories could match; 2) if the label is generated correctly; 3) if the mIOU is calculated correctly. Thank you so much.

[BUAA-LKG]

Basically, you can check: 1) if the colors (rgb) for the 11 categories could match; 2) if the label is generated correctly; 3) if the mIOU is calculated correctly. Thank you so much.

Ok, I will try. Another problem, what is the hardware environment when you train? 2 or 4 3090?

[XuJiacong]

Thanks. 2 A100 cards. I think 1 3090 is enough for camvid, but for cityscapes, I recommend at least 2 3090 cards.

[BUAA-LKG]

Thanks. 2 A100 cards. I think 1 3090 is enough for camvid, but for cityscapes, I recommend at least 2 3090 cards.

好的，我试一下复现！不知是否方面加个联系方式呢？我是北京航空航天大学的一名研究生，也在学习关于实时分割的方向！

[XuJiacong]

Please see the 'prepare the dataset' section in the front page of this github repository and check the method to download and prepare the data for camvid.

[XuJiacong]

谢谢关注， 你可以发你微信或者qq号到我的邮箱：[email protected]

[scl666]

谢谢关注， 你可以发你微信或者qq号到我的邮箱：[email protected]

min_value = pred[min(self.min_kept, pred.numel() - 1)] IndexError: index -1 is out of bounds for dimension 0 with size 0 你好，我在加载camvid数据集，出现这个错误。大佬训练时有遇到吗

[scl666]

Thanks. 2 A100 cards. I think 1 3090 is enough for camvid, but for cityscapes, I recommend at least 2 3090 cards.

好的，我试一下复现！不知是否方面加个联系方式呢？我是北京航空航天大学的一名研究生，也在学习关于实时分割的方向！

你好，这个camvid数据集你复现成功了吗？我使用这个数据集损失函数出错min_value = pred[min(self.min_kept, pred.numel() - 1)] IndexError: index -1 is out of bounds for dimension 0 with size 0，请问你有遇到吗

[Rebufleming]

Because I was curious how such high accuracy could be achieved on this dataset

你好！请问你复现到pid-s在camvid上的精度了吗？有些问题想请教，我的邮箱是[email protected]

[Rebufleming]

你好！请问你复现到pid-s在camvid上的精度了吗？有些问题想请教，我的邮箱是[email protected]

---

Thanks. 2 A100 cards. I think 1 3090 is enough for camvid, but for cityscapes, I recommend at least 2 3090 cards.

好的，我试一下复现！不知是否方面加个联系方式呢？我是北京航空航天大学的一名研究生，也在学习关于实时分割的方向！

你好，这个camvid数据集你复现成功了吗？我使用这个数据集损失函数出错min_value = pred[min(self.min_kept, pred.numel() - 1)] IndexError: index -1 is out of bounds for dimension 0 with size 0，请问你有遇到吗

你好！请问你复现到pid-s在camvid上的精度了吗？有些问题想请教，我的邮箱是[[email protected]]

[guodong110]

Thanks. 2 A100 cards. I think 1 3090 is enough for camvid, but for cityscapes, I recommend at least 2 3090 cards.

好的，我试一下复现！不知是否方面加个联系方式呢？我是北京航空航天大学的一名研究生，也在学习关于实时分割的方向！

你好，这个camvid数据集你复现成功了吗？我使用这个数据集损失函数出错min_value = pred[min(self.min_kept, pred.numel() - 1)] IndexError: index -1 is out of bounds for dimension 0 with size 0，请问你有遇到吗

您好，我在训练过程中也出现这个问题，请问是怎么解决的啊

[manuelrames]

Hi everybody! I also encountered the same problem when training a PIDNet small model on a custom dataset:

min_value = pred[min(self.min_kept, pred.numel() - 1)]
IndexError: index -1 is out of bounds for dimension 0 with size 0

Does anybody have any intuition about what would be the cause? Thanks in advance!

[manuelrames]

Hi everybody! I also encountered the same problem when training a PIDNet small model on a custom dataset:

min_value = pred[min(self.min_kept, pred.numel() - 1)]
IndexError: index -1 is out of bounds for dimension 0 with size 0

Does anybody have any intuition about what would be the cause? Thanks in advance!

Fixed it using the solution in https://github.com/XuJiacong/PIDNet/issues/14#issuecomment-1305105993

[17648145240]

谢谢。2 张 A100 卡。我认为 1 3090 对于 camvid 来说已经足够了，但对于城市景观，我建议至少 2 3090 张卡。

好的，我试一下复现！不知是否方面加个联系方式呢？我是北京航空航天大学的一名研究生，也在学习关于实时分割的方向！

你好，这个camvid数据集你复现成功了吗？我使用这个数据集损失函数出错min_value = pred[min（self.min_kept， pred.numel（） - 1）] IndexError： index -1 对于大小为 0 的维度 0 超出界限，请问你有遇到吗

兄弟，请问这个问题你解决了吗，我遇到了和你一样的问题，如果解决了的话，可以给我分享一下你修改之后的代码吗？我的邮箱：[email protected]

[wakakkakak]

谢谢关注， 你可以发你微信或者qq号到我的邮箱：[email protected]

min_value = pred[min(self.min_kept, pred.numel() - 1)] IndexError: index -1 is out of bounds for dimension 0 with size 0 你好，我在加载camvid数据集，出现这个错误。大佬训练时有遇到吗

您好我在复现时也遇到了同样的问题，请问您解决了吗，请问是如何修复的？（可以直接回复，也可以分享我的邮箱[email protected]）

[wakakkakak]

Thanks. 2 A100 cards. I think 1 3090 is enough for camvid, but for cityscapes, I recommend at least 2 3090 cards.

好的，我试一下复现！不知是否方面加个联系方式呢？我是北京航空航天大学的一名研究生，也在学习关于实时分割的方向！

你好，这个camvid数据集你复现成功了吗？我使用这个数据集损失函数出错min_value = pred[min(self.min_kept, pred.numel() - 1)] IndexError: index -1 is out of bounds for dimension 0 with size 0，请问你有遇到吗

您好，请问您解决了这个问题吗，是如何修改的呢？