请问是否可以使用其他预训练教师模型
Opened this issue · 3 comments
我用FCOS-GFL-R101的配置预训练了一个教师模型，并用教师训练了学生模型，基本达到了您论文的结果。
我想尝试使用其他的预训练模型训练教师，不知道可不可以，另外代码是不是需要修改某些部分，麻烦帮忙看一下，谢谢。
我对 LD/configs/ld/ld_r50_fcos_r101_1x.py 进行了更改，
teacher_config='configs/fcos/fcos_r101_caffe_fpn_gn-head_mstrain_640-800_2x_coco.py',
teacher_ckpt='configs/ld/fcos_r101_caffe_fpn_gn-head_mstrain_640-800_2x_coco-511424d6.pth'
并下载了mmdetection提供的预训练模型。
运行代码后出现了以下报错：
Traceback (most recent call last):
File "./tools/train.py", line 187, in
main()
File "./tools/train.py", line 183, in main
meta=meta)
File "/home/cx/LD/mmdet/apis/train.py", line 170, in train_detector
runner.run(data_loaders, cfg.workflow)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/epoch_based_runner.py", line 125, in run
epoch_runner(data_loaders[i], **kwargs)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/epoch_based_runner.py", line 50, in train
self.run_iter(data_batch, train_mode=True)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/epoch_based_runner.py", line 30, in run_iter
**kwargs)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/parallel/distributed.py", line 46, in train_step
Traceback (most recent call last):
File "./tools/train.py", line 187, in
Traceback (most recent call last):
main()
File "./tools/train.py", line 183, in main
File "./tools/train.py", line 187, in
meta=meta)
File "/home/cx/LD/mmdet/apis/train.py", line 170, in train_detector
runner.run(data_loaders, cfg.workflow)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/epoch_based_runner.py", line 125, in run
main()
File "./tools/train.py", line 183, in main
meta=meta)
File "/home/cx/LD/mmdet/apis/train.py", line 170, in train_detector
epoch_runner(data_loaders[i], **kwargs)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/epoch_based_runner.py", line 50, in train
runner.run(data_loaders, cfg.workflow)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/epoch_based_runner.py", line 125, in run
self.run_iter(data_batch, train_mode=True)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/epoch_based_runner.py", line 30, in run_iter
Traceback (most recent call last):
File "./tools/train.py", line 187, in
**kwargs)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/parallel/distributed.py", line 46, in train_step
main()
File "./tools/train.py", line 183, in main
meta=meta)
File "/home/cx/LD/mmdet/apis/train.py", line 170, in train_detector
runner.run(data_loaders, cfg.workflow)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/epoch_based_runner.py", line 125, in run
epoch_runner(data_loaders[i], **kwargs)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/epoch_based_runner.py", line 50, in train
epoch_runner(data_loaders[i], **kwargs)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/epoch_based_runner.py", line 50, in train
self.run_iter(data_batch, train_mode=True)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/epoch_based_runner.py", line 30, in run_iter
self.run_iter(data_batch, train_mode=True)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/epoch_based_runner.py", line 30, in run_iter
**kwargs)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/parallel/distributed.py", line 46, in train_step
**kwargs)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/parallel/distributed.py", line 46, in train_step
output = self.module.train_step(*inputs[0], **kwargs[0])
output = self.module.train_step(*inputs[0], **kwargs[0]) File "/home/cx/LD/mmdet/models/detectors/base.py", line 247, in train_step
File "/home/cx/LD/mmdet/models/detectors/base.py", line 247, in train_step
output = self.module.train_step(*inputs[0], **kwargs[0])
File "/home/cx/LD/mmdet/models/detectors/base.py", line 247, in train_step
output = self.module.train_step(*inputs[0], **kwargs[0])
File "/home/cx/LD/mmdet/models/detectors/base.py", line 247, in train_step
losses = self(**data)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
losses = self(**data)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
losses = self(**data)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
losses = self(**data)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/fp16_utils.py", line 84, in new_func
result = self.forward(*input, **kwargs)result = self.forward(*input, **kwargs)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/fp16_utils.py", line 84, in new_func
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/fp16_utils.py", line 84, in new_func
result = self.forward(*input, **kwargs)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/fp16_utils.py", line 84, in new_func
return old_func(*args, **kwargs)
File "/home/cx/LD/mmdet/models/detectors/base.py", line 181, in forward
return old_func(*args, **kwargs)
return old_func(*args, **kwargs)
File "/home/cx/LD/mmdet/models/detectors/base.py", line 181, in forward
File "/home/cx/LD/mmdet/models/detectors/base.py", line 181, in forward
return old_func(*args, **kwargs)
File "/home/cx/LD/mmdet/models/detectors/base.py", line 181, in forward
return self.forward_train(img, img_metas, **kwargs)
File "/home/cx/LD/mmdet/models/detectors/kd_one_stage.py", line 76, in forward_train
return self.forward_train(img, img_metas, **kwargs)
return self.forward_train(img, img_metas, **kwargs) File "/home/cx/LD/mmdet/models/detectors/kd_one_stage.py", line 76, in forward_train
File "/home/cx/LD/mmdet/models/detectors/kd_one_stage.py", line 76, in forward_train
gt_bboxes_ignore)
File "/home/cx/LD/mmdet/models/dense_heads/ld_fcos_head.py", line 254, in forward_train
gt_bboxes_ignore)
File "/home/cx/LD/mmdet/models/dense_heads/ld_fcos_head.py", line 254, in forward_train
gt_bboxes_ignore)
File "/home/cx/LD/mmdet/models/dense_heads/ld_fcos_head.py", line 254, in forward_train
losses = self.loss(*loss_inputs, gt_bboxes_ignore=gt_bboxes_ignore)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/fp16_utils.py", line 164, in new_func
losses = self.loss(*loss_inputs, gt_bboxes_ignore=gt_bboxes_ignore)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/fp16_utils.py", line 164, in new_func
losses = self.loss(*loss_inputs, gt_bboxes_ignore=gt_bboxes_ignore)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/fp16_utils.py", line 164, in new_func
return self.forward_train(img, img_metas, **kwargs)
File "/home/cx/LD/mmdet/models/detectors/kd_one_stage.py", line 76, in forward_train
return old_func(*args, **kwargs)
File "/home/cx/LD/mmdet/models/dense_heads/ld_fcos_head.py", line 204, in loss
return old_func(*args, **kwargs)
File "/home/cx/LD/mmdet/models/dense_heads/ld_fcos_head.py", line 204, in loss
return old_func(*args, **kwargs)num_pos=num_pos)
File "/home/cx/LD/mmdet/models/dense_heads/ld_fcos_head.py", line 204, in loss
File "/home/cx/LD/mmdet/core/utils/misc.py", line 29, in multi_apply
num_pos=num_pos)
File "/home/cx/LD/mmdet/core/utils/misc.py", line 29, in multi_apply
gt_bboxes_ignore)
File "/home/cx/LD/mmdet/models/dense_heads/ld_fcos_head.py", line 254, in forward_train
num_pos=num_pos)
File "/home/cx/LD/mmdet/core/utils/misc.py", line 29, in multi_apply
losses = self.loss(*loss_inputs, gt_bboxes_ignore=gt_bboxes_ignore)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/mmcv/runner/fp16_utils.py", line 164, in new_func
return old_func(*args, **kwargs)
File "/home/cx/LD/mmdet/models/dense_heads/ld_fcos_head.py", line 204, in loss
num_pos=num_pos)
File "/home/cx/LD/mmdet/core/utils/misc.py", line 29, in multi_apply
return tuple(map(list, zip(*map_results)))return tuple(map(list, zip(*map_results)))
return tuple(map(list, zip(*map_results)))
File "/home/cx/LD/mmdet/models/dense_heads/ld_fcos_head.py", line 57, in loss_single
File "/home/cx/LD/mmdet/models/dense_heads/ld_fcos_head.py", line 57, in loss_single
File "/home/cx/LD/mmdet/models/dense_heads/ld_fcos_head.py", line 57, in loss_single
return tuple(map(list, zip(*map_results)))
File "/home/cx/LD/mmdet/models/dense_heads/ld_fcos_head.py", line 57, in loss_single
4 * (self.reg_max + 1))4 * (self.reg_max + 1))
4 * (self.reg_max + 1))
RuntimeError: shape '[-1, 68]' is invalid for input of size 243200RuntimeError
: shape '[-1, 68]' is invalid for input of size 243200
RuntimeError: shape '[-1, 68]' is invalid for input of size 268800
4 * (self.reg_max + 1))
RuntimeError: shape '[-1, 68]' is invalid for input of size 243200
Traceback (most recent call last):
File "/home/.conda/envs/cx_ld/lib/python3.7/runpy.py", line 193, in _run_module_as_main
"main", mod_spec)
File "/home/.conda/envs/cx_ld/lib/python3.7/runpy.py", line 85, in _run_code
exec(code, run_globals)
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/torch/distributed/launch.py", line 260, in
main()
File "/home/.conda/envs/cx_ld/lib/python3.7/site-packages/torch/distributed/launch.py", line 256, in main
cmd=cmd)
configs/fcos/fcos_r101_caffe_fpn_gn-head_mstrain_640-800_2x_coco.py 和 FCOS-GFL-R101
区别在于一个用FCOSHead，另一个用FCOSGFLHead，这两个除了一般分布没有任何区别
不改成一般分布是没办法用LD的