a  d@stddlZddlmZddlmZddlmZeeejejedddZ eeeedd d Z deeeedd d Z dS)N)Tensor)conv2d)reduce)win_sizesigmadtypedevicereturncCs^tj|||d|dd}|d}t|d|d d|d}|t|}|S)Nrrr@)torchZarangeexp unsqueezesum)rrrrZcoordsgrj/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/torchmetrics/functional/image/vif.py_filters (r)predstarget sigma_n_sqr cCsV|j}|j}|d}|d}tjd||d}tj|||d}tjd||dtjd||d}}tdD]}dd|d} t| | d||dddddf} |dkrt|| ddddddddddf}t|| ddddddddddf}t|| } t|| } | d} | d}| | }tj t|d| | d d }tj t|d| |d d }t||| |}|||}|||}||k}d||<||||<d||<||k}d||<d||<|dk}||||<d||<tj ||d }t d |d|||}|tj |gd d }|tj t d ||gd d }qj||S)Nr g|=r r rr g)ming?)r r )Zdim) rrrrZtensorzerosrangerrclamplog10r)rrrrrZepsZ preds_vifZ target_vifZscalenkernelZ mu_targetZmu_predsZ mu_target_sqZ mu_preds_sqZmu_target_predsZsigma_target_sqZsigma_preds_sqZsigma_target_predsrZ sigma_v_sqmaskZpreds_vif_scalerrr_vif_per_channel!sL  "" **      &r$r csddksddkr>tdddddddksZddkr|tdddddfdd td D}tt|d S) aCompute Pixel Based Visual Information Fidelity (VIF_). Args: preds: predicted images of shape ``(N,C,H,W)``. ``(H, W)`` has to be at least ``(41, 41)``. target: ground truth images of shape ``(N,C,H,W)``. ``(H, W)`` has to be at least ``(41, 41)`` sigma_n_sq: variance of the visual noise Return: Tensor with vif-p score Raises: ValueError: If ``data_range`` is neither a ``tuple`` nor a ``float`` )z8Invalid size of preds. Expected at least 41x41, but got x!z9Invalid size of target. Expected at least 41x41, but got c sHg|]@}tdd|ddddfdd|ddddfqS)N)r$).0irrrrr rz/visual_information_fidelity..r Zelementwise_mean)size ValueErrorrrrcat)rrrZ per_channelrr,rvisual_information_fidelityVs" r2)r ) rrZtorch.nn.functionalrZ"torchmetrics.utilities.distributedrfloatrrrr$r2rrrrs     5