The model and loaded state dict do not match exactly when trying to load pretrained weights

[Chris Hughes]

Hi,

When I am trying to run a training epoch using pretrained weights, I am getting a warning that the model and state dict do not match exactly. As I have not modified anything, I suspect that it is a weight/config mismatch.

Whilst there is an example for convnext_tiny, I am making some assumptions here as I am trying to train convnext_base, the command I am using is:
python tools/train.py configs/convnext/cascade_mask_rcnn_convnext_base_patch4_window7_mstrain_480-800_giou_4conv1f_adamw_3x_coco_in1k.py --cfg-options model.pretrained=https://dl.fbaipublicfiles.com/convnext/coco/cascade_mask_rcnn_convnext_base_1k_3x.pth

Is this the correct pretrained model for this config?

Thanks!

[wszhengjx]

have you resolved it ? i meet the same question

[Chris Hughes]

Unfortunately I was never able to resolve it, so abandoned investigating it further as it doesn't look actively maintained