Wrong layers (names?) when extracting all conv layers from AlexNet.
Closed this issue · 4 comments
spaladin commented
Hi!
I am using the Collab notebook for PyTorch.
I tried Alexnet – by changing some parameters in the notebook examples (VGG-16 with batch norm pretrained on ImageNet)
I wanted to extract activations from all the convolutional layers, so I used the following code modification:
pretrained = True
model_path = None
batch_size = 32
apply_center_crop = True
flatten_activations = True
class_names = None
file_names = None
file_format = "txt"
device = 'cuda' if torch.cuda.is_available() else 'cpu'model_name = 'alexnet'
extractor = get_extractor(
model_name=model_name,
pretrained=pretrained,
model_path=model_path,
device=device,
source=source
)layer = nn.Conv2d
features_conv_layers = extract_all_layers(
model_name=model_name,
extractor=extractor,
image_path=full_image_path,
out_path=full_output_path,
batch_size=batch_size,
flatten_activations=flatten_activations,
apply_center_crop=apply_center_crop,
layer=layer,
file_format=file_format,
class_names=class_names,
file_names=file_names,
)The number of extracted layers (5) is correct, however, their names are wrong. Also, they don’t seem to match activations if I extract them layer by layer.
import pprint
pprint.pprint(features_conv_layers)> {'layer_03': array([[ 0.12578183, -1.6219227 , -1.70176 , ..., -0.09369432,
> 0.08984605, 0.53728145],
> [-0.7437832 , 0.4840992 , -1.0749123 , ..., -0.5906651 ,
> -0.0103681 , 0.85010654],
> [-7.1971655 , -7.940407 , -6.6123657 , ..., -0.03713347,
> 0.17685306, 0.20328899],
> ...,
> [-0.11900061, 0.45805246, -0.24691838, ..., 0.43640044,
> 0.60965 , 0.15608627],
> [-3.3795178 , -0.8164063 , 1.241787 , ..., -0.15670502,
> 0.37275872, -1.1117904 ],
> [-1.8591139 , 0.42655402, -1.2627366 , ..., 0.13208538,
> 0.7406758 , -0.08220328]], dtype=float32),
> 'layer_06': array([[ 0.9620508 , 0.30162707, -0.1934116 , ..., 1.858568 ,
> 1.1076739 , 2.6782532 ],
> [ -2.981957 , -2.7508926 , 2.3697217 , ..., -3.5808048 ,
> -3.423941 , -0.64591765],
> [ 4.2801642 , -0.83435506, 2.985959 , ..., 6.7334747 ,
> 6.2881317 , 5.9027047 ],
> ...,
> [ 1.0700345 , 3.1900558 , 10.337312 , ..., -15.291744 ,
> -13.26463 , -8.463119 ],
> [ 4.113041 , -0.5327738 , -7.635279 , ..., -13.555077 ,
> -11.066146 , -6.3253107 ],
> [ 7.7316403 , 4.5360703 , 2.8130543 , ..., -1.4696633 ,
> -1.934596 , -2.111609 ]], dtype=float32),
> 'layer_09': array([[ -2.7296655 , -3.8743975 , -4.42631 , ..., -1.769564 ,
> -9.306692 , -3.7036123 ],
> [ -0.04922827, 8.507316 , 8.73079 , ..., -1.6609691 ,
> -2.1132905 , 1.4741383 ],
> [ 1.260469 , 2.5003388 , 6.413993 , ..., -13.279748 ,
> -11.688034 , -0.75490063],
> ...,
> [ -5.667928 , -4.167725 , -3.8533154 , ..., -5.4969683 ,
> -5.595151 , -4.3243084 ],
> [-13.667816 , -12.076218 , -13.503252 , ..., -9.637596 ,
> -10.82891 , -7.7699695 ],
> [ -3.5512943 , -6.506268 , -8.565321 , ..., -8.342162 ,
> -5.3187623 , -3.7092364 ]], dtype=float32),
> 'layer_11': array([[ -1.8959136 , -0.49840185, 1.8974366 , ..., -0.53857136,
> -2.0079157 , -0.99538684],
> [ -6.2139473 , -6.6683593 , 0.59373176, ..., -2.49711 ,
> -0.7762851 , 5.3251452 ],
> [ 0.98107946, 2.1276023 , 0.21201959, ..., -5.2047086 ,
> -2.7088084 , -3.1409965 ],
> ...,
> [ 0.8351224 , -2.8610208 , -0.6101372 , ..., -7.19401 ,
> -6.187467 , -3.8016884 ],
> [ -0.98556584, 3.609405 , 5.1768856 , ..., -12.374811 ,
> -10.80249 , -9.2378645 ],
> [ -1.0655197 , -2.2648356 , 0.3327253 , ..., -1.8608872 ,
> -1.464742 , 2.7857661 ]], dtype=float32),
> 'layer_13': array([[-5.751322 , -7.13288 , -7.124266 , ..., -5.815998 , -5.2186804,
> -1.9163358],
> [-3.6160367, -5.2578096, -4.3458047, ..., -6.0499434, -5.955589 ,
> -3.924894 ],
> [-1.8869516, -8.714518 , -6.626626 , ..., 0.0618896, -6.0121365,
> -7.5943756],
> ...,
> [-1.597339 , -2.036482 , -2.903429 , ..., -8.830565 , -8.896645 ,
> -8.055616 ],
> [-2.4018128, -5.483816 , -5.364626 , ..., -7.1737504, -6.7403235,
> -4.4299064],
> [-6.647602 , -5.406938 , -3.681465 , ..., -3.7315614, -4.8392906,
> -4.518252 ]], dtype=float32)}
LukasMut commented
Hi there! This is because the keys for the features_per_layer dictionary that is used to store the activations for each module in the extract_all_layers convenience function (see below) correspond to custom names and not to the original module names. You can replace features_per_layer[f'layer_{l:02d}'] with features_per_layer[f'{module_name}'] for using original module names. See my comments in the below function.
def extract_all_layers(
model_name: str,
extractor: Any,
image_path: str,
out_path: str,
batch_size: int,
flatten_activations: bool,
apply_center_crop: bool,
layer: Any=nn.Linear,
file_format: str = "npy",
class_names: Optional[List[str]]=None,
file_names: Optional[List[str]]=None,
) -> Dict[str, Union[np.ndarray, torch.Tensor]]:
"""Extract features for all selected layers and save them to disk."""
features_per_layer = {}
for l, (module_name, module) in enumerate(extractor.model.named_modules(), start=1):
if isinstance(module, layer):
# extract features for layer "module_name"
features = extract_features(
extractor=extractor,
module_name=module_name,
image_path=image_path,
out_path=out_path,
batch_size=batch_size,
flatten_activations=flatten_activations,
apply_center_crop=apply_center_crop,
class_names=class_names,
file_names=file_names,
)
# NOTE: for original module names use features_per_layer[f'{module_name}'] = features
# replace with e.g., [f'conv_{l:02d}'] or [f'fc_{l:02d}']
features_per_layer[f'layer_{l:02d}'] = features
# save the features to disk
save_features(features, out_path=f'{out_path}/features_{model_name}_{module_name}', file_format=file_format)
return features_per_layerspaladin commented
Hi Lukas! Thanks for the quick response!
I guess, I am missing something, but after replacing the line as suggested, I get only one key named 'module_name' like this:
> {'module_name': array([[-5.751324 , -7.1328826, -7.124266 , ..., -5.815997 , -5.2186813,
> -1.916337 ],
> [-3.6160486, -5.257819 , -4.3458114, ..., -6.049942 , -5.9555883,
> -3.9248927],
> [-1.8869545, -8.714518 , -6.6266246, ..., 0.0618932, -6.012134 ,
> -7.594375 ],
> ...,
> [-1.5973396, -2.0364807, -2.9034288, ..., -8.830564 , -8.896644 ,
> -8.055617 ],
> [-2.4018128, -5.4838195, -5.3646264, ..., -7.173752 , -6.7403235,
> -4.4299088],
> [-6.647603 , -5.406933 , -3.6814663, ..., -3.7315552, -4.8392887,
> -4.5182486]], dtype=float32)}
LukasMut commented
Hi Lukas! Thanks for the quick response!
I guess, I am missing something, but after replacing the line as suggested, I get only one key named 'module_name' like this:
> {'module_name': array([[-5.751324 , -7.1328826, -7.124266 , ..., -5.815997 , -5.2186813, > -1.916337 ], > [-3.6160486, -5.257819 , -4.3458114, ..., -6.049942 , -5.9555883, > -3.9248927], > [-1.8869545, -8.714518 , -6.6266246, ..., 0.0618932, -6.012134 , > -7.594375 ], > ..., > [-1.5973396, -2.0364807, -2.9034288, ..., -8.830564 , -8.896644 , > -8.055617 ], > [-2.4018128, -5.4838195, -5.3646264, ..., -7.173752 , -6.7403235, > -4.4299088], > [-6.647603 , -5.406933 , -3.6814663, ..., -3.7315552, -4.8392887, > -4.5182486]], dtype=float32)}
I am sorry. It should read features_per_layer[f'{module_name}'] rather than features_per_layer[f'module_name']. This was just a typo in my previous comment (I've corrected it).
spaladin commented
Great, thank you, that works!