yuhuan-wu
commented
4 months ago Hi Genbao,
Thank you for your email. I understand your issue.
The logs I released were trained using an earlier version of the mmdet library. Since mmdet has updated the way it loads pretrained models, there may be inconsistencies as you mentioned.
To check if the pretrained model has been successfully loaded, please search for logs that indicate "loading checkpoints/pretrained weights" or similar. You can conveniently search using the model file's path.
Additionally, you can compare the speed of loss reduction in my logs with your training logs. If the loss in your training is decreasing significantly slower than in my logs, it is likely that the pretrained model was not loaded by the mmdet library code.
Best Regards, Yu-Huan
Genbao-Xu
Hello author, I have a question about loading pretrained models for object detection tasks. The following is the part of loading the pre-trained model in my training log file, which is a little different from the result of loading the pretrained model in the log file you provided (yours shows 'pretrained=.... ' directly after type='retinanet', is there any difference between the two, and whether I successfully loaded the pretrained model, thank you.) My: model = dict( type='RetinaNet', backbone=dict( type='p2t_tiny', depth=50， num stages=4, out indices=(0，1，2，3), frozen stages=1, norm cfg=dict(type='BN', requires_grad=True) norm eval=True, style='pytorch', init_cfg=dict( type='Pretrained', checkpoint='pretrained/p2t_tiny.pth' ), )
Yours: model = dict(
type='RetinaNet',
pretrained='data/pretrained/p2t/p2t_tiny.pth', backbone=dict( type='p2t_tiny', depth=50, num_stages=4, out_indices=(0, 1, 2, 3), frozen_stages=1, norm_cfg=dict(type='BN', requires_grad=True), norm_eval=True, style='pytorch' )