640 resolution

[liangxuejingjing]

Thank you for releasing this excellent project, I wanna run it on high resolution, e.g. 2048*2048, how can I change the network?

[lyuwenyu]

Which codebase (torch or paddle) do you use?

Just find and modify *eval_size in related config file.

---

You can star this repo to follow updates, https://github.com/lyuwenyu/RT-DETR/issues/179

[liangxuejingjing]

Which codebase (torch or paddle) do you use?

Just find and modify *eval_size in related config file.

You can star this repo to follow updates, #179

torch

[liangxuejingjing]

Which codebase (torch or paddle) do you use?

Just find and modify *eval_size in related config file.

You can star this repo to follow updates, #179

Though I have run rt-detr with higher resolution， the output is not ideal. How can I add a feature map “S6”?

[lyuwenyu]

try to modify num_levels in decoder.

https://github.com/lyuwenyu/RT-DETR/blob/main/rtdetr_pytorch/configs/rtdetr/include/rtdetr_r50vd.yml#L50

[dakdouky]

Which codebase (torch or paddle) do you use?

Just find and modify *eval_size in related config file.

You can star this repo to follow updates, #179

Hi @lyuwenyu,

Does this require retraining, or can this be just done for inference?

[chhchhchh]

为什么修改完https://github.com/lyuwenyu/RT-DETR/blob/main/rtdetr_pytorch/configs/rtdetr/include/rtdetr_r50vd.yml#L50配置文件后，还是只有3层特征层的输出？