assert (boxes1[:, 2:] >= boxes1[:, :2]).all() AssertionError

[chenqq1]

assert (boxes1[:, 2:] >= boxes1[:, :2]).all() AssertionError

[ShoufaChen]

Hi, I didn't meet this error in our experiments.

Did you use coco dataset or your custom data?

[chenqqqqqqq]

Hi, I didn't meet this error in our experiments.

Did you use coco dataset or your custom data?

我用的是自己定义的数据集，这个报错在以下两种情况下都有发生：

1. 当我将AdamW优化器更换为一个最新的Lion优化器时这个错误会出现；
2. 我使用AdamW优化器，backbone选用resnet50时是可以对自己定义的数据集进行完整训练的。可是，当我将backbone换成resnet101时，迭代到第两万次时，这个错误又出现了！ 不懂到底是为什么，难道是学习率每次都要做不同的调整吗？因为我的数据集成功跑通过其它很多模型，所以数据集是没有问题的。希望您可以给我解答一下。

[asanc199]

Hi! I'm having the same problem, did you managed to solve it?

[chenqqqqqqq]

Sorry, i didn't sovle it. ---- Replied Message ---- | From | Alejandro Sanchez @.> | | Date | 07/02/2023 23:58 | | To | ShoufaChen/DiffusionDet @.> | | Cc | chenqqqqqqq @.>, Comment @.> | | Subject | Re: [ShoufaChen/DiffusionDet] assert (boxes1[:, 2:] >= boxes1[:, :2]).all() AssertionError (Issue #60) |

Hi! I'm having the same problem, did you managed to solve it?

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

[asanc199]

Hi! Thank you chenq, i realized that i could handle it by changing de optimizer to SGD instead of using the default AdamW. Hope it helps

[chenqqqqqqq]

yeah, I realize that this error is impacted by the optimizer. When i use AdamW, it runs normally, but when i change AdamW with other optimizer, such as the latest optimizer Lion, it goes wrong.

| | 陈萌 | | @.*** | 签名由网易邮箱大师定制 On 7/3/2023 17:18，Alejandro Sanchez @.***> wrote：

Hi! Thank you chenq, i realized that i could handle it by changing de optimizer to SGD instead of using the default AdamW. Hope it helps

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>