Summary:
Users have to explicitly opt-in for those transforms. Here we provide the first building block for this interface. We add the functionals `to_nvcv_image` and `nvcv_to_tensor` to transform `torch.Tensor` to `nvcv.Tensor`. We also implement the corresponding class transforms `ToNVCVImage` and `NVCVToTensor`.

## How to use

```python
from PIL import Image
import torchvision.transforms.v2.functional as F

orig_img = Image.open("leaning_tower.jpg")
img_tensor = F.pil_to_tensor(orig_img)
nvcv_tensor = F.to_nvcv_tensor(img_tensor.cuda())
img_tensor = F.nvcv_to_tensor(nvcv_tensor)
```

> [!NOTE]
> NVCV tensors are automatically converted to NHWC format. Contrary to torchvision convention, which relies on NCHW format.

## Run unit tests

```bash
pytest test/test_cvcuda.py
...
37 passed in 0.15s
```


Test Plan:
```python
from torchvision import _is_cvcuda_available

_is_cvcuda_available()
```

## Run tests

```bash
buck test fbcode//mode/opt fbcode//pytorch/vision/test:torchvision_cvcuda
...
Tests finished: Pass 38. Fail 0. Fatal 0. Skip 0. Build failure 0
```

Differential Revision: D85862362

Pulled By: AntoineSimoulin

Summary:
Users have to explicitly opt-in for those transforms. Here we provide the first building block for this interface. We add the functionals `to_nvcv_image` and `nvcv_to_tensor` to transform `torch.Tensor` to `nvcv.Tensor`. We also implement the corresponding class transforms `ToNVCVImage` and `NVCVToTensor`.

## How to use

```python
from PIL import Image
import torchvision.transforms.v2.functional as F

orig_img = Image.open("leaning_tower.jpg")
img_tensor = F.pil_to_tensor(orig_img)
nvcv_tensor = F.to_nvcv_tensor(img_tensor.cuda())
img_tensor = F.nvcv_to_tensor(nvcv_tensor)
```

> [!NOTE]
> NVCV tensors are automatically converted to NHWC format. Contrary to torchvision convention, which relies on NCHW format.

## Run unit tests

```bash
pytest test/test_cvcuda.py
...
37 passed in 0.15s
```


Test Plan:
```python
from torchvision import _is_cvcuda_available

_is_cvcuda_available()
```

## Run tests

```bash
buck test fbcode//mode/opt fbcode//pytorch/vision/test:torchvision_cvcuda
...
Tests finished: Pass 38. Fail 0. Fatal 0. Skip 0. Build failure 0
```

Differential Revision: D85862362

Pulled By: AntoineSimoulin

Summary:
Summary
-------

This PR provides the first building blocks for CV-CUDA integration in torchvision. We add the functionals `to_cvcuda_tensor` and `cvcuda_to_tensor` to transform from `torch.Tensor` to `cvcuda.Tensor` and back. We also implement the corresponding class transforms `ToCVCUDATensor` and `CVCUDAToTensor`.

**Key features:**

*   **3-channel RGB support only**: Simplified API focusing on the most common use case (RGB images)
*   **Supported data types**: `torch.uint8` (RGB8 format) and `torch.float32` (RGBf32 format)
*   **Lossless round-trip conversions**: Exact data preservation when converting PyTorch ↔ CV-CUDA
*   **Informative error messages**: Helpful installation instructions when CV-CUDA is not available
*   **Batch-aware**: Handles both unbatched (CHW) and batched (NCHW) tensors

Users must explicitly opt-in for these transforms, which require CV-CUDA to be installed.

How to use
----------

```python
from PIL import Image
import torchvision.transforms.v2.functional as F 

# Load and convert image to PyTorch tensor
orig_img = Image.open("leaning_tower.jpg")
img_tensor = F.pil_to_tensor(orig_img)  

# Convert to CV-CUDA tensor (must be 3-channel RGB on CUDA)
cvcuda_tensor = F.to_cvcuda_tensor(img_tensor.cuda())

# Convert back to PyTorch tensor
img_tensor = F.cvcuda_to_tensor(cvcuda_tensor)
```

> [!NOTE]
> 
> *   NVCV tensors are automatically converted to NHWC layout, contrary to torchvision's NCHW default
> *   Only 3-channel RGB images and 1-channel grayscale are supported for now
> *   Input tensors will be uploaded to CUDA device when converting to CV-CUDA tensors
> *   CV-CUDA must be installed: `pip install cvcuda-cu12` (CUDA 12) or `pip install cvcuda-cu11` (CUDA 11)

Run unit tests
--------------
## Run unit tests

```bash
pytest test/test_transforms_v2.py -k "cvcuda"
...
35 passed, 4 skipped, 9774 deselected in 1.12s
```


Test Plan:
```python
from torchvision import _is_cvcuda_available

_is_cvcuda_available()
```

## Run tests

```bash
buck test fbcode//mode/opt fbcode//pytorch/vision/test:torchvision_transforms_v2
...
Tests finished: Pass 38. Fail 0. Fatal 0. Skip 0. Build failure 0
```

Differential Revision: D85862362

Pulled By: AntoineSimoulin

Summary:
Summary
-------

This PR provides the first building blocks for CV-CUDA integration in torchvision. We add the functionals `to_cvcuda_tensor` and `cvcuda_to_tensor` to transform from `torch.Tensor` to `cvcuda.Tensor` and back. We also implement the corresponding class transforms `ToCVCUDATensor` and `CVCUDAToTensor`.

**Key features:**

*   **3-channel RGB support only**: Simplified API focusing on the most common use case (RGB images)
*   **Supported data types**: `torch.uint8` (RGB8 format) and `torch.float32` (RGBf32 format)
*   **Lossless round-trip conversions**: Exact data preservation when converting PyTorch ↔ CV-CUDA
*   **Informative error messages**: Helpful installation instructions when CV-CUDA is not available
*   **Batch-aware**: Handles both unbatched (CHW) and batched (NCHW) tensors

Users must explicitly opt-in for these transforms, which require CV-CUDA to be installed.

How to use
----------

```python
from PIL import Image
import torchvision.transforms.v2.functional as F 

# Load and convert image to PyTorch tensor
orig_img = Image.open("leaning_tower.jpg")
img_tensor = F.pil_to_tensor(orig_img)  

# Convert to CV-CUDA tensor (must be 3-channel RGB on CUDA)
cvcuda_tensor = F.to_cvcuda_tensor(img_tensor.cuda())

# Convert back to PyTorch tensor
img_tensor = F.cvcuda_to_tensor(cvcuda_tensor)
```

> [!NOTE]
> 
> *   NVCV tensors are automatically converted to NHWC layout, contrary to torchvision's NCHW default
> *   Only 3-channel RGB images and 1-channel grayscale are supported for now
> *   Input tensors will be uploaded to CUDA device when converting to CV-CUDA tensors
> *   CV-CUDA must be installed: `pip install cvcuda-cu12` (CUDA 12) or `pip install cvcuda-cu11` (CUDA 11)

Run unit tests
--------------

```bash
pytest test/test_transforms_v2.py -k "cvcuda"
...
35 passed, 4 skipped, 9774 deselected in 1.12s
```


Differential Revision: D85862362

Pulled By: AntoineSimoulin