I’m new to PyTorch, and I am just trying to use the pretrained Mask R-CNN model to segment a single image. To do this, I’m reading the image using img=PIL.Image.open("apple.jpg")
. Now I simply need to transform the image to a tensor to pass to the model, but how can I do this in the simplest way possible? I can use torchvision.transforms.transforms.Compose([transforms.ToTensor()])(img)
, but is it not possible to apply the ToTensor
transform without using Compose
?
You don’t even need to use pil.
You can use any image reader you want imageio, opencv, scikit-image (all those returns numpy arrays)
and just
img = load(...)
img = torch.from_numpy(img)/255.
and there you go
or alternatevely you can do transforms.ToTensor()(img)
1 Like