行为识别 * 双流卷积网络

 

  * Two stream average fusion at softmax layer.  - （在原双流论文中）时间和空间两个神经网络的融合是在最后一步，结果取平均值或者用线性SVM划 分

'''
using pre-trained weights on ImageNet for both streams
'''
spat_model = models.vgg16(pretrained=True)
temp_model = models.vgg16(pretrained=True)




if use_gpu:
    spat_model = spat_model.cuda()
    temp_model = temp_model.cuda()

spat_criterion = nn.CrossEntropyLoss()
temp_criterion = nn.CrossEntropyLoss()




Code borrowed from：tomar840/two-stream-fusion-for-action-recognition-in-videos