Resnet101 worse than Resnet50?

[chenqqqqqqq]

Experiment on a custom defective dataset. Why is the detection accuracy worse on Resnet101 than on Resnet50? Is it related to the learning rate?

[ShoufaChen]

Hi @chenqqqqqqq ,

Thanks for your interest.

We compared DiffusionDet with res50 and res100 on coco and lvis datasets and the results demonstrate resnet101 is better.

In your case, I guess it is an overfitting phenomenon. What's your custom dataset size?

[chenqqqqqqq]

我在NEU-DET数据集上训练的，一共包含1800幅灰度图像，图像尺寸为200×200。

| | 陈萌 | | @.*** | 签名由网易邮箱大师定制 On 5/11/2023 16:43，Shoufa @.***> wrote：

Hi @chenqqqqqqq ,

Thanks for your interest.

We compared DiffusionDet with res50 and res100 on coco and lvis datasets and the results demonstrate resnet101 is better.

In your case, I guess it is an overfitting phenomenon. What's your custom dataset size?

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