StandardSensor

class dnois.sensor.StandardSensor(pixel_num: int | tuple[int, int], pixel_size: float | tuple[float, float], rgb: bool = True, srf: Tensor | tuple[Tensor, Tensor, Tensor] = None, bayer_pattern: Literal['RGGB', 'GRBG', 'BGGR', 'GBRG'] = None, noise_std: float | tuple[float, float] = 0.0, max_value: float = 1.0, _quantize: int = 256, linear2srgb: bool = True, differentiable_quant: bool = True)

A simple RGB or grayscale sensor model, which processes the radiance field reaching the sensor plane as follows:

  1. Divide pixels into channels according to Bayer CFA.

  2. Spectral integral by srf.

  3. Apply additional Gaussian white noise.

  4. Restrict signal values to a given range.

  5. Quantize signal to a given level.

Parameters:

  • pixel_num (int or tuple[int, int]) – Numbers of pixels in vertical and horizontal directions.

  • pixel_size (float or tuple[float, float]) – Height and width of a pixel in meters.

  • rgb (bool) – RGB sensor if True, grayscale sensor otherwise. Default: True.

  • srf (Tensor or tuple[Tensor, Tensor, Tensor]) – SRF tensor of shape (N_C, N_wl) where N_C is 1 or 3, or a 3-tuple of tensors of length N_wl, corresponding to the SRF of R, G, B channels. Default: do not perform spectral integral.

  • bayer_pattern (BayerPattern) – See rgb2raw(). Default: no CFA.

  • noise_std (float or tuple[float, float]) – Standard deviation of the Gaussian white noise. See gaussian(). Default: 0.

  • max_value (float) – Maximum possible value of output signal. Default: 1.

  • _quantize (int) – Quantization level. The output signal will not be quantized if a negative or zero value is given. Default: 256.

  • linear2srgb (bool) – Whether to convert linear RGB to sRGB. Default: True.

  • differentiable_quant (bool) – See quantize(). Default: True.

extra_repr() → str

Return the extra representation of the module.

To print customized extra information, you should re-implement this method in your own modules. Both single-line and multi-line strings are acceptable.

forward(radiance: Tensor) → Tensor

Simulate the process of conversion from optical signal radiance to electric signal, an 2D image.

Parameters:

radiance (Tensor) – A tensor representing the received radiance field, of shape (..., N_wl, H, W).

Returns:

Output image signal, a tensor of shape (..., N_C', H, W). If self.srf is not None, radiance is monochromatic (N_wl=1) or self.bayer_pattern is not None, N_C' is 1. Otherwise, N_C' is 3 (RGB regardless of SRF and Bayer CFA).

Return type:

Tensor

bayer_pattern: BayerPattern | None

Bayer CFA pattern. See rgb2raw()

differentiable_quantization: bool

Whether to perform differentiable quantization.

property h

Physical height of the sensor in meters.

Type:

float

linear2srgb: bool

Whether to convert linear RGB to sRGB.

max_value: float

Maximum possible value of signal.

noise_std: float | tuple[float, float]

Standard deviation of Gaussian noise.

property pixel_h: float

Height of a pixel.

Type:

float

pixel_num: tuple[int, int]

Numbers of pixels in vertical and horizontal directions.

pixel_size: tuple[float, float]

Height and width of a pixel in meters.

property pixel_w: float

Width of a pixel.

Type:

float

quantize: int

Quantization level.

property res_h: int

Number of pixels in vertical direction.

Type:

int

property res_w: int

Number of pixels in horizontal direction.

Type:

int

rgb: bool

RGB sensor or not.

property size: tuple[float, float]

Returns the physical size i.e. height and width of the sensor.

Type:

tuple[float, float]

srf: Tensor

Spectral response functions. Either None or a tensor of shape (1, N_wl) or (4, N_wl) (two identical green SRF in Bayer CFA)

property w

Physical width of the sensor in meters.

Type:

float