Convolutional layers in ResNet152 are freezed in stage 1 about openlongtailrecognition-oltr

[RamyaRaghuraman]

I am also facing the same issue during stage_1 training:

RuntimeError:
An attempt has been made to start a new process before the
current process has finished its bootstrapping phase.

    This probably means that you are not using fork to start your
    child processes and you have forgotten to use the proper idiom
    in the main module:

        if __name__ == '__main__':
            freeze_support()
            ...

    The "freeze_support()" line can be omitted if the program
    is not going to be frozen to produce an executable.

[zhmiao]

@jchhuang @RamyaRaghuraman Actually, as written in the paper, we used frozen resnet 152 feature extractor for the places experiments. The first reason is that finetuning resnet 152 requires large gpu resource, which we don't have. Second, as I explained in #22 (comment) self attention module and meta embedding module require initialization. Since the feature extractor is already imagenet pretrained, so self attention module won't be affected too much. However, in stage 1, we are still training classifier module with an additional fc layer. The memory is initialized using the feature after the fc, and the meta embedding was initialized using the stage 1 pretrained classifier. So the training of stage 1 is not meaningless.

[RamyaRaghuraman]

@zhmiao Thank you for your reply. I understand that the stage 1 training is required, but I am still not able to overcome the stated "freeze_support()" error. Is there a work around for this?

[zhmiao]

@RamyaRaghuraman Could you be a little bit more specific on the error?

[RamyaRaghuraman]

@zhmiao I have attached the entire output for stage_1 training. I am using only 1 GPU but I do not know where the multiprocessing error comes from.

`C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\python.exe C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/main.py
Loading dataset from: C:\Users\RAR7ABT\Documents\Meine empfangenen Dateien\Places_LT
{'criterions': {'PerformanceLoss': {'def_file': 'C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/loss/SoftmaxLoss.py',
'loss_params': {},
'optim_params': None,
'weight': 1.0}},
'memory': {'centroids': False, 'init_centroids': False},
'networks': {'classifier': {'def_file': 'C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/models/DotProductClassifier.py',
'optim_params': {'lr': 0.1,
'momentum': 0.9,
'weight_decay': 0.0005},
'params': {'dataset': 'Places_LT',
'in_dim': 512,
'num_classes': 365,
'stage1_weights': False}},
'feat_model': {'def_file': 'C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/models/ResNet152Feature.py',
'fix': True,
'optim_params': {'lr': 0.01,
'momentum': 0.9,
'weight_decay': 0.0005},
'params': {'caffe': True,
'dataset': 'Places_LT',
'dropout': None,
'stage1_weights': False,
'use_fc': True,
'use_modulatedatt': False}}},
'training_opt': {'batch_size': 256,
'dataset': 'Places_LT',
'display_step': 10,
'feature_dim': 512,
'log_dir': 'C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/logs/Places_LT/stage1',
'num_classes': 365,
'num_epochs': 30,
'num_workers': 4,
'open_threshold': 0.1,
'sampler': None,
'scheduler_params': {'gamma': 0.1, 'step_size': 10}}}
Loading data from C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/data/Places_LT/Places_LT_train.txt
Use data transformation: Compose(
RandomResizedCrop(size=(224, 224), scale=(0.08, 1.0), ratio=(0.75, 1.3333), interpolation=PIL.Image.BILINEAR)
RandomHorizontalFlip(p=0.5)
ColorJitter(brightness=[0.6, 1.4], contrast=[0.6, 1.4], saturation=[0.6, 1.4], hue=None)
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)
txt C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/data/Places_LT/Places_LT_train.txt
No sampler.
Shuffle is True.
Loading data from C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/data/Places_LT/Places_LT_val.txt
Use data transformation: Compose(
Resize(size=256, interpolation=PIL.Image.BILINEAR)
CenterCrop(size=(224, 224))
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)
txt C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/data/Places_LT/Places_LT_val.txt
No sampler.
Shuffle is True.
cuda:0
Using 1 GPUs.
Loading Scratch ResNet 152 Feature Model.
Using fc.
Loading Caffe Pretrained ResNet 152 Weights.
Pretrained feature model weights path: C:\Users\RAR7ABT\pj-val-ml\pjval_ml\OSR\OLTR\logs\resnet152.pth
Freezing feature weights except for self attention weights (if exist).
Loading Dot Product Classifier.
Random initialized classifier weights.
Using steps for training.
Initializing model optimizer.
Loading Softmax Loss.
Phase: train
Loading dataset from: C:\Users\RAR7ABT\Documents\Meine empfangenen Dateien\Places_LT
{'criterions': {'PerformanceLoss': {'def_file': 'C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/loss/SoftmaxLoss.py',
'loss_params': {},
'optim_params': None,
'weight': 1.0}},
'memory': {'centroids': False, 'init_centroids': False},
'networks': {'classifier': {'def_file': 'C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/models/DotProductClassifier.py',
'optim_params': {'lr': 0.1,
'momentum': 0.9,
'weight_decay': 0.0005},
'params': {'dataset': 'Places_LT',
'in_dim': 512,
'num_classes': 365,
'stage1_weights': False}},
'feat_model': {'def_file': 'C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/models/ResNet152Feature.py',
'fix': True,
'optim_params': {'lr': 0.01,
'momentum': 0.9,
'weight_decay': 0.0005},
'params': {'caffe': True,
'dataset': 'Places_LT',
'dropout': None,
'stage1_weights': False,
'use_fc': True,
'use_modulatedatt': False}}},
'training_opt': {'batch_size': 256,
'dataset': 'Places_LT',
'display_step': 10,
'feature_dim': 512,
'log_dir': 'C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/logs/Places_LT/stage1',
'num_classes': 365,
'num_epochs': 30,
'num_workers': 4,
'open_threshold': 0.1,
'sampler': None,
'scheduler_params': {'gamma': 0.1, 'step_size': 10}}}
Loading data from C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/data/Places_LT/Places_LT_train.txt
Use data transformation: Compose(
RandomResizedCrop(size=(224, 224), scale=(0.08, 1.0), ratio=(0.75, 1.3333), interpolation=PIL.Image.BILINEAR)
RandomHorizontalFlip(p=0.5)
ColorJitter(brightness=[0.6, 1.4], contrast=[0.6, 1.4], saturation=[0.6, 1.4], hue=None)
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)
txt C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/data/Places_LT/Places_LT_train.txt
No sampler.
Shuffle is True.
Loading data from C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/data/Places_LT/Places_LT_val.txt
Use data transformation: Compose(
Resize(size=256, interpolation=PIL.Image.BILINEAR)
CenterCrop(size=(224, 224))
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)
txt C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/data/Places_LT/Places_LT_val.txt
No sampler.
Shuffle is True.
cuda:0
Using 1 GPUs.
Loading Scratch ResNet 152 Feature Model.
Using fc.
Loading Caffe Pretrained ResNet 152 Weights.
Pretrained feature model weights path: C:\Users\RAR7ABT\pj-val-ml\pjval_ml\OSR\OLTR\logs\resnet152.pth
Freezing feature weights except for self attention weights (if exist).
Loading Dot Product Classifier.
Random initialized classifier weights.
Using steps for training.
Initializing model optimizer.
Loading Softmax Loss.
Phase: train
Traceback (most recent call last):
File "", line 1, in
Traceback (most recent call last):
File "C:/Users/RAR7ABT/pj-val-ml/pjval_ml/OSR/OLTR/main.py", line 60, in
training_model.train()
File "C:\Users\RAR7ABT\pj-val-ml\pjval_ml\OSR\OLTR\run_networks.py", line 208, in train
for step, (inputs, labels, _) in enumerate(self.data['train']):
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\site-packages\torch\utils\data\dataloader.py", line 193, in iter
return _DataLoaderIter(self)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\site-packages\torch\utils\data\dataloader.py", line 469, in init
w.start()
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\process.py", line 105, in start
self._popen = self._Popen(self)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\context.py", line 223, in _Popen
return _default_context.get_context().Process._Popen(process_obj)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\context.py", line 322, in _Popen
return Popen(process_obj)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\popen_spawn_win32.py", line 65, in init
reduction.dump(process_obj, to_child)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\reduction.py", line 60, in dump
ForkingPickler(file, protocol).dump(obj)
BrokenPipeError: [Errno 32] Broken pipe
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\spawn.py", line 105, in spawn_main
exitcode = _main(fd)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\spawn.py", line 114, in _main
prepare(preparation_data)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\spawn.py", line 225, in prepare
_fixup_main_from_path(data['init_main_from_path'])
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\spawn.py", line 277, in _fixup_main_from_path
run_name="mp_main")
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\runpy.py", line 263, in run_path
pkg_name=pkg_name, script_name=fname)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\runpy.py", line 96, in _run_module_code
mod_name, mod_spec, pkg_name, script_name)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\runpy.py", line 85, in _run_code
exec(code, run_globals)
File "C:\Users\RAR7ABT\pj-val-ml\pjval_ml\OSR\OLTR\main.py", line 60, in
training_model.train()
File "C:\Users\RAR7ABT\pj-val-ml\pjval_ml\OSR\OLTR\run_networks.py", line 208, in train
for step, (inputs, labels, _) in enumerate(self.data['train']):
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\site-packages\torch\utils\data\dataloader.py", line 193, in iter
return _DataLoaderIter(self)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\site-packages\torch\utils\data\dataloader.py", line 469, in init
w.start()
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\process.py", line 105, in start
self._popen = self._Popen(self)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\context.py", line 223, in _Popen
return _default_context.get_context().Process._Popen(process_obj)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\context.py", line 322, in _Popen
return Popen(process_obj)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\popen_spawn_win32.py", line 33, in init
prep_data = spawn.get_preparation_data(process_obj._name)
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\spawn.py", line 143, in get_preparation_data
_check_not_importing_main()
File "C:\Users\RAR7ABT\AppData\Local\conda\conda\envs\pjval\lib\multiprocessing\spawn.py", line 136, in _check_not_importing_main
is not going to be frozen to produce an executable.''')
RuntimeError:
An attempt has been made to start a new process before the
current process has finished its bootstrapping phase.

    This probably means that you are not using fork to start your
    child processes and you have forgotten to use the proper idiom
    in the main module:

        if __name__ == '__main__':
            freeze_support()
            ...

    The "freeze_support()" line can be omitted if the program
    is not going to be frozen to produce an executable.

Process finished with exit code 1
`

[zhmiao]

@RamyaRaghuraman Thanks for posting the error message. It seems to be a problem of running pytorch on windows instead of frozen weights. Here is an issue from pytorch official repo: pytorch/pytorch#5858 (comment) . I will close this issue and open a new one about this (#29 (comment)). We do not have a windows machine to test it for now. I think we will try to get one and see if it is possible to reproduce. Thanks.

[jchhuang]

@jchhuang @RamyaRaghuraman Actually, as written in the paper, we used frozen resnet 152 feature extractor for the places experiments. The first reason is that finetuning resnet 152 requires large gpu resource, which we don't have. Second, as I explained in #22 (comment) self attention module and meta embedding module require initialization. Since the feature extractor is already imagenet pretrained, so self attention module won't be affected too much. However, in stage 1, we are still training classifier module with an additional fc layer. The memory is initialized using the feature after the fc, and the meta embedding was initialized using the stage 1 pretrained classifier. So the training of stage 1 is not meaningless.

Hi, I don't agree this explain. In my understanding, the dimension of centroid is num_classnum_feature, however, the output of fc is a 1num_class vector, their size are not matched, how can a vector initialize a matrix？