i just set with_mask = False and modified some relavant codes to just do bbox detection without segm.It's works when i just train a model from scratch,but when i try to use the command here:
python train.py mask_rcnn_swin_tiny_patch4_window7_mstrain_480-800_adamw_1x_coco.py --resume-from mask_rcnn_swin_tiny_patch4_window7_1x.pth.
it reports error:
Traceback (most recent call last):
File "D:\dev_proj\Swin-Transformer-Object-Detection-master\train.py", line 188, in
main()
File "D:\dev_proj\Swin-Transformer-Object-Detection-master\train.py", line 177, in main
train_detector(
File "D:\dev_proj\Swin-Transformer-Object-Detection-master\mmdet\apis\train.py", line 182, in train_detector
runner.resume(cfg.resume_from)
File "D:\dev_proj\Swin-Transformer-Object-Detection-master\mmcv_custom\runner\epoch_based_runner.py", line 86, in resume
self._epoch = checkpoint['meta']['epoch']
~~~~~~~~~~~~~~~~~~^^^^^^^^^
KeyError: 'epoch'

seems like i cant use a pretrained model because i just delete some mask head module in the model,can anyone tell how to fix it so that i can use a pretrained model to improve my model performance.A model without pretrained perfomes really bad in my dataset.