Why are there NaN values in extracted features?

vision
1

Hi

I am extracting features from a video using ResNet152 from torchvision, but when examining the features I found that some contain nan values. I tried using Densenet161 but got the same thing.

That’s how I define the model

class ResNet:
    resnets = {
        50: resnet.resnet50,
        152: resnet.resnet152
    }

    def __init__(self, i):
        model = ResNet.resnets[i](pretrained=True)
        model = nn.Sequential(*list(model.children())[:-1])
        model.eval()
        model.cuda()
        self.model = model
        
    def __call__(self, x):
        out = self.model(x)
        return out

and I preprocess each frame in the video accordingly

normalize = transforms.Normalize(
     mean=[0.485, 0.456, 0.406],
     std=[0.229, 0.224, 0.225]
)

preprocess = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    normalize
])

any idea how this can be solved?

1 Like