o oi@sdZddlmZddlmZddlZddlmZmZm Z ddl m Z m Z ddl mZd%d d Zd&d'ddZd(d)ddZd*d+ddZd(d,d#d$ZdS)-zImplementation of "differentiable spatial to numerical" (soft-argmax) operations. As described in the paper "Numerical Coordinate Regression with Convolutional Neural Networks" by Nibali et al. ) annotations)OptionalN)Tensor concatenatesoftmaxKORNIA_CHECK_IS_TENSORKORNIA_CHECK_SHAPE)create_meshgridtensorrreturnNonecCst|t|gddS)N)BCHWr)r rO/home/ubuntu/.local/lib/python3.10/site-packages/kornia/geometry/subpix/dsnt.py$_validate_batched_image_tensor_input"srinput temperatureOptional[Tensor]cCsht||j\}}}}|durtd}|j|j|jd}|||d}t||dd}|||||S)a2Apply the Softmax function over features in each image channel. Note that this function behaves differently to :py:class:`torch.nn.Softmax2d`, which instead applies Softmax over features at each spatial location. Args: input: the input tensor with shape :math:`(B, N, H, W)`. temperature: factor to apply to input, adjusting the "smoothness" of the output distribution. Returns: a 2D probability distribution per image channel with shape :math:`(B, N, H, W)`. Examples: >>> heatmaps = torch.tensor([[[ ... [0., 0., 0.], ... [0., 0., 0.], ... [0., 1., 2.]]]]) >>> spatial_softmax2d(heatmaps) tensor([[[[0.0585, 0.0585, 0.0585], [0.0585, 0.0585, 0.0585], [0.0585, 0.1589, 0.4319]]]]) Ng?)devicedtype)dim) rshapetorchr torrviewr)rr batch_sizechannelsheightwidthxx_softrrrspatial_softmax2d's r&Tnormalized_coordinatesboolc Cst||j\}}}}t||||j}||j}|dd}|dd}|||d} tj || ddd} tj || ddd} t | | gd} | ||dS)aCompute the expectation of coordinate values using spatial probabilities. The input heatmap is assumed to represent a valid spatial probability distribution, which can be achieved using :func:`~kornia.geometry.subpixel.spatial_softmax2d`. Args: input: the input tensor representing dense spatial probabilities with shape :math:`(B, N, H, W)`. normalized_coordinates: whether to return the coordinates normalized in the range of :math:`[-1, 1]`. Otherwise, it will return the coordinates in the range of the input shape. Returns: expected value of the 2D coordinates with shape :math:`(B, N, 2)`. Output order of the coordinates is (x, y). Examples: >>> heatmaps = torch.tensor([[[ ... [0., 0., 0.], ... [0., 0., 0.], ... [0., 1., 0.]]]]) >>> spatial_expectation2d(heatmaps, False) tensor([[[1., 2.]]]) .rr.Tkeepdim) rrr rrrreshaperrsumr) rr'r r!r"r#gridpos_xpos_y input_flat expected_y expected_xoutputrrrspatial_expectation2dLs r83#I9 numerator denominatorepsfloatcCs|tj||dS)N)min)rclamp)r:r;r<rrr_safe_zero_divisionysr@meanstdsizetuple[int, int]cCs|j|jkr |j|jkstd|\}}t||||j}||j}|d||}|d||}||dd} ||dd} dt|d} dt|d} t| | } t| | }| |}|j dd d j d d d }t ||}|S) aRender the PDF of a 2D Gaussian distribution. Args: mean: the mean location of the Gaussian to render, :math:`(\mu_x, \mu_y)`. Shape: :math:`(*, 2)`. std: the standard deviation of the Gaussian to render, :math:`(\sigma_x, \sigma_y)`. Shape :math:`(*, 2)`. Should be able to be broadcast with `mean`. size: the (height, width) of the output image. normalized_coordinates: whether ``mean`` and ``std`` are assumed to use coordinates normalized in the range of :math:`[-1, 1]`. Otherwise, coordinates are assumed to be in the range of the output shape. Returns: tensor including rendered points with shape :math:`(*, H, W)`. z1Expected inputs to have the same dtype and devicer)r*).rNNr.).r+NNgTr,r) rr TypeErrorr rrr reciprocalexpr0r@)rArBrCr'r"r#r1r2r3dist_xdist_yk_xk_yexps_xexps_ygaussval_sumrrrrender_gaussian2d}s"  rQ)r rr r )N)rrrrr r)T)rrr'r(r r)r9)r:rr;rr<r=r r) rArrBrrCrDr'r(r r)__doc__ __future__rtypingrr kornia.corerrrkornia.core.checkrr kornia.utils.gridr rr&r8r@rQrrrrs      % -