Why time steps is increasing?

I am training a pytorch model and I remark that the time steps is increasing as mentioned below:

I remark as well that my GPU is 0% use

Here is the transform that I am using:

def get_transform(config):
    scale = torchvision.transforms.Lambda(lambda x: x.astype(np.float32) / 255.)
    identity = torchvision.transforms.Lambda(lambda x: x)
    size = config.transform.mpiifacegaze_face_size
    if size != 448:
        resize = torchvision.transforms.Resize((size, size))
        resize = identity
    if config.transform.mpiifacegaze_gray:
        to_gray = torchvision.transforms.Lambda(lambda x: cv2.cvtColor(
            cv2.equalizeHist(cv2.cvtColor(x, cv2.COLOR_BGR2GRAY)), cv2.
        to_gray = identity

    transform = torchvision.transforms.Compose([
        torchvision.transforms.Lambda(lambda x: np.transpose(x, [2, 0, 1])),
        torchvision.transforms.Normalize(mean=[0.406, 0.456, 0.485],
                                         std=[0.225, 0.224, 0.229]),
    return transform

Are you only applying the posted transformation in your code snippet without any model training?
If so, then I would expect the GPU utilization to be at 0%, since the CPU will be used.
For the increase in iteration time: I guess you might be storing some tensors, which are still attached to the computation graph, and are indeed training the model. This would increase the memory usage as well as the processing time.

No, I am training a deep model based on mobilenet.