Closed l-dawei closed 9 months ago
Hello.
Is the teacher_pretrained used in the second training the same as the ckpt obtained in the first train stage?
Who can help me?
# method details model = dict( _delete_ = True, type='PoseEstimatorDistiller', two_dis = second_dis, teacher_pretrained = 'work_dirs/rtmpose_x_dis_l__coco-ubody-256x192/dw-x-l_ucoco.pth', teacher_cfg = 'configs/wholebody_2d_keypoint/rtmpose/ubody/rtmpose-l_8xb64-270e_coco-ubody-wholebody-256x192.py', student_cfg = 'configs/wholebody_2d_keypoint/rtmpose/ubody/rtmpose-l_8xb64-270e_coco-ubody-wholebody-256x192.py', distill_cfg = [ dict(methods=[dict(type='KDLoss', name='loss_logit', use_this = logit, weight = 1, ) ] ), ], data_preprocessor=dict( type='PoseDataPreprocessor', mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], bgr_to_rgb=True), train_cfg=train_cfg, )
Yes, you can use the ckpt obtained from the first distillation stage or the baseline training. Both two models are fine.
Hello.
Is the teacher_pretrained used in the second training the same as the ckpt obtained in the first train stage?
Who can help me?