Training log update about fgd HOT 6 CLOSED

The log cannot help you to solve the problem. There must be something wrong with the codes. What's your mmdet version?

from fgd.

The log cannot help you to solve the problem. There must be something wrong with the codes. What's your mmdet version?

mmdet==2.25.0 mmcv-full==1.6.1

from fgd.

Please refer README for the higher mmdet version. You might initialize the student and teacher wrongly.

from fgd.

In my training log, I found
2022-08-05 17:52:24,038 - mmdet - INFO - Set random seed to 0, deterministic: False
2022-08-05 17:52:26,558 - mmdet - INFO - initialize ResNeXt with init_cfg {'type': 'Pretrained', 'checkpoint': 'open-mmlab://resnext101_64x4d'}
2022-08-05 17:52:27,039 - mmdet - INFO - initialize FPN with init_cfg {'type': 'Xavier', 'layer': 'Conv2d', 'distribution': 'uniform'}
2022-08-05 17:52:27,095 - mmdet - INFO - initialize RetinaHead with init_cfg {'type': 'Normal', 'layer': 'Conv2d', 'std': 0.01, 'override': {'type': 'Normal', 'name': 'retina_cls', 'std': 0.01, 'bias_prob': 0.01}}
2022-08-05 17:52:27,234 - mmdet - INFO - initialize ResNet with init_cfg {'type': 'Pretrained', 'checkpoint': 'torchvision://resnet50'}
2022-08-05 17:52:27,452 - mmdet - INFO - initialize FPN with init_cfg {'type': 'Xavier', 'layer': 'Conv2d', 'distribution': 'uniform'}
2022-08-05 17:52:27,509 - mmdet - INFO - initialize RetinaHead with init_cfg {'type': 'Normal', 'layer': 'Conv2d', 'std': 0.01, 'override': {'type': 'Normal', 'name': 'retina_cls', 'std': 0.01, 'bias_prob': 0.01}}
It seems that student and teacher all init from pretrained model, right?

from fgd.

For mmdet>2.12, you have to change the tools/train.py and mmdet/distillation/distillers/detection_distiller.py. You can refer MGD to change this.

from fgd.

OK, I will change code refer MGD.
Thanks for your reply!!

from fgd.