ttnn.rotate

ttnn.rotate(input_tensor: ttnn.Tensor, angle: float, *, center: Tuple[float, float] | None, fill: float, expand: bool, interpolation_mode: str, memory_config: ttnn.MemoryConfig | None) → ttnn.Tensor

Rotates a tensor by an arbitrary angle around a specified center point using configurable interpolation.

The rotate operation performs spatial transformation by rotating each pixel position around a rotation center. Areas outside the rotated tensor are filled with a configurable fill value.

Parameters:

  • input_tensor (ttnn.Tensor) – Input tensor of shape (N, H, W, C) in NHWC format

  • angle (float) – Rotation angle in degrees. Positive values rotate counter-clockwise

Keyword Arguments:

  • center (Optional[Tuple[float, float]]) – Rotation center point as (x, y) in pixel coordinates, where x is the horizontal/width coordinate and y is the vertical/height coordinate. Default: tensor center at ((W-1)/2, (H-1)/2)

  • fill (float) – Fill value for areas outside the rotated tensor. Default: 0.0

  • expand (bool) – If True, return error. Only False is supported (same output dimensions). Default: False

  • interpolation_mode (str) – Interpolation method - “nearest” (sharp, faster) or “bilinear” (smooth, slower). Default: “nearest”

  • memory_config (ttnn.MemoryConfig, optional) – Output memory configuration. Default: DRAM_INTERLEAVED

Returns:

ttnn.Tensor – Rotated tensor of shape (N, H, W, C) - same as input

Example

>>> # Create input tensor (N=1, H=256, W=256, C=32) - channel last format
>>> input_tensor = ttnn.from_torch(torch.randn(1, 256, 256, 32), device=device)
>>>
>>> # Rotate 45 degrees counter-clockwise with nearest interpolation (default)
>>> output = ttnn.rotate(input_tensor, 45.0)
>>> print(output.shape)  # [1, 256, 256, 32]
>>>
>>> # Rotate 90 degrees clockwise with nearest interpolation (faster)
>>> output_cw = ttnn.rotate(input_tensor, -90.0, interpolation_mode="nearest")
>>>
>>> # Rotate with bilinear interpolation (smoother but slower)
>>> output_smooth = ttnn.rotate(input_tensor, 45.0, interpolation_mode="bilinear")
>>>
>>> # Rotate around custom center (x=128, y=64) with white fill
>>> output_custom = ttnn.rotate(input_tensor, 30.0, center=(128, 64), fill=1.0)