如何从预先训练的PyTorch模型(例如ResNet或VGG)的特定层提取特征,而无需再次进行前向传递?
最佳答案
您可以在所需的特定层上注册forward hook。就像是:
def some_specific_layer_hook(module, input_, output):
pass # the value is in 'output'
model.some_specific_layer.register_forward_hook(some_specific_layer_hook)
model(some_input)
例如,要在ResNet中获得res5c输出,您可能需要使用
nonlocal
变量(或Python 2中的global
):res5c_output = None
def res5c_hook(module, input_, output):
nonlocal res5c_output
res5c_output = output
resnet.layer4.register_forward_hook(res5c_hook)
resnet(some_input)
# Then, use `res5c_output`.