Changing torchvision namespace into a static class?

[kaiidams]

Quite a few code starts from torchvision instead of importing objects by from torchvision... import xxx

https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html

model = torchvision.models.detection.fasterrcnn_resnet50_fpn(weights="DEFAULT")

Now in TorchSharp, we have TorchSharp.torchvision namespace.

namespace TorchSharp.torchvision
{
    public static partial class models
    {

Can we make this a class TorchSharp.torchvision like we did for TorchSharp.torch class?

namespace TorchSharp
{
    public static partial class torchvision
    {
        public static partial class models
        {

I believe that we have to do like these in C#

using TorchSharp.torchvision;
...
models.detection.fasterrcnn_resnet50_fpn();

or

using static TorchSharp.torchvision.models.detection;
...
fasterrcnn_resnet50_fpn();

[NiklasGustafsson]

Any time a named scope directly contains methods, or is nested within a scope that contains methods, whether instance factories or other, it needs to be a static class. If that is the case with 'torchvision,' it should definitely be a static class.

Thus, it depends on what methods would be found directly under the 'torchvision' named scope.

[kaiidams]

PyTorch torchvision has four methods

• torchvision.set_image_backend(backend)
• torchvision.get_image_backend()
• torchvision.set_video_backend(backend)
• torchvision.get_video_backend()

TorchSharp has Imager as an argument of torchvision.io.read_image() instead of having those four methods.
So currently TorchSharp doesn't have any methods under the torchvision named scope.

A namespace has to start from the top. Currently we cannot write like this in the user code.

using TorchSharp;
...
torchvision.io.read_image()