Retrieving the logits for the class predictions for the Faster RCNN model in torchvision

Question

I have this code to predict an image using FASTER RCNN model:

    import torch
    import torchvision
    from torchvision import transforms as T

    model = torchvision.models.detection.fasterrcnn_resnet50_fpn(pretrained = True)

    model.eval()

    img = Image.open("../image.jpg")

    transform = T.Compose([T.ToTensor()])

    ig = transform(img).unsqueeze(0)

    with torch.no_grad():
    pred = model(ig)
 

    The result of pred like that: 
    [{'boxes': tensor([[3.9041e+00, 3.3453e+02, 1.4249e+02, 4.0109e+02],
          [7.1592e+02, 2.9045e+02, 1.0998e+03, 5.9793e+02],
          [5.9441e+02, 2.3850e+02, 7.3092e+02, 5.3254e+02],]),
     'labels': tensor([ 3,  3,  1]),
     'scores': tensor([0.9970, 0.9961, 0.9961])}]

I want to retrieve the 'logits' for the class predictions for the Faster RCNN model in   torchvision, 
["logits" refers to the raw, unnormalized scores or probabilities assigned to different classes   by a model. Logits are the output of the last layer of a neural network before it undergoes a normalization operation (often softmax) to produce probabilities.]

How I can do that ? 

I have tried to do like that:

pred['logits'] but I got this error: 'Keyerror: 'logits''.