Bad results with pretrained SiamRPN models???

[researchmm]

Bad results with pretrained SiamRPN models???

[JudasDie]

Hi, Zhipeng, good job! Thanks for your kind release of this code. When I test the siamRPN with pre-trained model, however, I find the results is always lost even at the beginning. I wonder if you released the right version of the pretrained model? I see the following on my terminal:

python test_siamrpn.py
load pretrained model from ../snapshot/CIResNet22.pth
remove prefix 'module.'
missing keys:set(['connect_model.adjust.weight', 'connect_model.search_reg.bias', 'connect_model.search_cls.bias', 'connect_model.search_reg.weight', 'connect_model.template_reg.weight', 'connect_model.template_reg.bias', 'connect_model.search_cls.weight', 'connect_model.template_cls.bias', 'connect_model.template_cls.weight', 'connect_model.adjust.bias'])
unused checkpoint keys:set([u'connect_model.loc_adjust.weight', u'connect_model.loc_adjust.bias'])

Would you please kindly help to run the code succesfully? the code of SiamFC is OK (I tried). I used this setting for testing:

parser.add_argument('--arch', dest='arch', default='SiamRPNRes22', help='backbone architecture')
parser.add_argument('--resume', default='../snapshot/CIResNet22.pth', type=str, help='pretrained model')

Hi~ The pretrained model you used is not correct. You should use CIResNet22_RPN.pth for SiamRPN and CIResNet22.pth for SiamFC respectively. Try and give me a feedback pls.

[wangxiao5791509]

Thanks for your advice. I also tried "CIResNet22_RPN.pth", but it shown me the following error:

python test_siamrpn.py
load pretrained model from ../snapshot/CIResNet22_RPN.pth
remove prefix 'module.'
missing keys:set([])
unused checkpoint keys:set([u'features.features.layer2.2.bn1.num_batches_tracked', u'features.features.layer2.0.bn2.num_batches_tracked', u'features.features.layer2.3.bn2.num_batches_tracked', u'features.features.layer1.2.bn1.num_batches_tracked', u'features.features.layer1.1.bn3.num_batches_tracked', u'features.features.layer1.2.bn2.num_batches_tracked', u'features.features.layer2.4.bn2.num_batches_tracked', u'features.features.layer2.3.bn3.num_batches_tracked', u'features.features.layer1.0.bn1.num_batches_tracked', u'features.features.layer2.3.bn1.num_batches_tracked', u'features.features.layer2.2.bn3.num_batches_tracked', u'features.features.layer1.1.bn2.num_batches_tracked', u'features.features.layer2.0.bn3.num_batches_tracked', u'features.features.layer1.1.bn1.num_batches_tracked', u'features.features.layer2.2.bn2.num_batches_tracked', u'features.features.layer1.2.bn3.num_batches_tracked', u'features.features.layer1.0.downsample.1.num_batches_tracked', u'features.features.layer2.0.downsample.1.num_batches_tracked', u'features.features.layer2.4.bn3.num_batches_tracked', u'features.features.layer1.0.bn2.num_batches_tracked', u'features.features.bn1.num_batches_tracked', u'features.features.layer2.0.bn1.num_batches_tracked', u'features.features.layer2.4.bn1.num_batches_tracked', u'features.features.layer1.0.bn3.num_batches_tracked'])
Traceback (most recent call last):
File "test_siamrpn.py", line 317, in
main()
File "test_siamrpn.py", line 203, in main
net = load_pretrain(net, args.resume)
File "/media/wangxiao/49cd8079-e619-4e4b-89b1-15c86afb5102/my_works/SiamDW-master-revised/siamese_tracking/../lib/utils/utils.py", line 240, in load_pretrain
model.load_state_dict(pretrained_dict, strict=False)
File "/usr/local/lib/python2.7/dist-packages/torch/nn/modules/module.py", line 721, in load_state_dict
self.class.name, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for SiamRPNRes22:
While copying the parameter named "connect_model.template_cls.bias", whose dimensions in the model are torch.Size([5120]) and whose dimensions in the checkpoint are torch.Size([1280]).
While copying the parameter named "connect_model.template_cls.weight", whose dimensions in the model are torch.Size([5120, 512, 3, 3]) and whose dimensions in the checkpoint are torch.Size([1280, 512, 3, 3]).
While copying the parameter named "connect_model.template_reg.bias", whose dimensions in the model are torch.Size([10240]) and whose dimensions in the checkpoint are torch.Size([5120]).
While copying the parameter named "connect_model.template_reg.weight", whose dimensions in the model are torch.Size([10240, 512, 3, 3]) and whose dimensions in the checkpoint are torch.Size([5120, 512, 3, 3]).
While copying the parameter named "connect_model.search_cls.bias", whose dimensions in the model are torch.Size([512]) and whose dimensions in the checkpoint are torch.Size([256]).
While copying the parameter named "connect_model.search_cls.weight", whose dimensions in the model are torch.Size([512, 512, 3, 3]) and whose dimensions in the checkpoint are torch.Size([256, 512, 3, 3]).
While copying the parameter named "connect_model.search_reg.bias", whose dimensions in the model are torch.Size([512]) and whose dimensions in the checkpoint are torch.Size([256]).
While copying the parameter named "connect_model.search_reg.weight", whose dimensions in the model are torch.Size([512, 512, 3, 3]) and whose dimensions in the checkpoint are torch.Size([256, 512, 3, 3]).

[JudasDie]

Thanks for your advice. I also tried "CIResNet22_RPN.pth", but it shown me the following error:

python test_siamrpn.py
load pretrained model from ../snapshot/CIResNet22_RPN.pth
remove prefix 'module.'
missing keys:set([])
unused checkpoint keys:set([u'features.features.layer2.2.bn1.num_batches_tracked', u'features.features.layer2.0.bn2.num_batches_tracked', u'features.features.layer2.3.bn2.num_batches_tracked', u'features.features.layer1.2.bn1.num_batches_tracked', u'features.features.layer1.1.bn3.num_batches_tracked', u'features.features.layer1.2.bn2.num_batches_tracked', u'features.features.layer2.4.bn2.num_batches_tracked', u'features.features.layer2.3.bn3.num_batches_tracked', u'features.features.layer1.0.bn1.num_batches_tracked', u'features.features.layer2.3.bn1.num_batches_tracked', u'features.features.layer2.2.bn3.num_batches_tracked', u'features.features.layer1.1.bn2.num_batches_tracked', u'features.features.layer2.0.bn3.num_batches_tracked', u'features.features.layer1.1.bn1.num_batches_tracked', u'features.features.layer2.2.bn2.num_batches_tracked', u'features.features.layer1.2.bn3.num_batches_tracked', u'features.features.layer1.0.downsample.1.num_batches_tracked', u'features.features.layer2.0.downsample.1.num_batches_tracked', u'features.features.layer2.4.bn3.num_batches_tracked', u'features.features.layer1.0.bn2.num_batches_tracked', u'features.features.bn1.num_batches_tracked', u'features.features.layer2.0.bn1.num_batches_tracked', u'features.features.layer2.4.bn1.num_batches_tracked', u'features.features.layer1.0.bn3.num_batches_tracked'])
Traceback (most recent call last):
File "test_siamrpn.py", line 317, in
main()
File "test_siamrpn.py", line 203, in main
net = load_pretrain(net, args.resume)
File "/media/wangxiao/49cd8079-e619-4e4b-89b1-15c86afb5102/my_works/SiamDW-master-revised/siamese_tracking/../lib/utils/utils.py", line 240, in load_pretrain
model.load_state_dict(pretrained_dict, strict=False)
File "/usr/local/lib/python2.7/dist-packages/torch/nn/modules/module.py", line 721, in load_state_dict
self.class.name, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for SiamRPNRes22:
While copying the parameter named "connect_model.template_cls.bias", whose dimensions in the model are torch.Size([5120]) and whose dimensions in the checkpoint are torch.Size([1280]).
While copying the parameter named "connect_model.template_cls.weight", whose dimensions in the model are torch.Size([5120, 512, 3, 3]) and whose dimensions in the checkpoint are torch.Size([1280, 512, 3, 3]).
While copying the parameter named "connect_model.template_reg.bias", whose dimensions in the model are torch.Size([10240]) and whose dimensions in the checkpoint are torch.Size([5120]).
While copying the parameter named "connect_model.template_reg.weight", whose dimensions in the model are torch.Size([10240, 512, 3, 3]) and whose dimensions in the checkpoint are torch.Size([5120, 512, 3, 3]).
While copying the parameter named "connect_model.search_cls.bias", whose dimensions in the model are torch.Size([512]) and whose dimensions in the checkpoint are torch.Size([256]).
While copying the parameter named "connect_model.search_cls.weight", whose dimensions in the model are torch.Size([512, 512, 3, 3]) and whose dimensions in the checkpoint are torch.Size([256, 512, 3, 3]).
While copying the parameter named "connect_model.search_reg.bias", whose dimensions in the model are torch.Size([512]) and whose dimensions in the checkpoint are torch.Size([256]).
While copying the parameter named "connect_model.search_reg.weight", whose dimensions in the model are torch.Size([512, 512, 3, 3]) and whose dimensions in the checkpoint are torch.Size([256, 512, 3, 3]).

1. Please install packages with install_rpn.sh.
2. Make sure that the input cls_type is thinner.

[wangxiao5791509]

@JudasDie Thanks for your suggestions. It's working now (Only "thinner" works for my setting). May I add your wechat? Maybe we can have some cooperations later. It's ok if you do not want to be troubled. My wechat is: wangxiao5791509.

[JudasDie]

@JudasDie Thanks for your suggestions. It's working now (Only "thinner" works for my setting). May I add your wechat? Maybe we can have some cooperations later. It's ok if you do not want to be troubled. My wechat is: wangxiao5791509.

Glad to know you make it. The question will be closed.