a dD@@sddlZddlZddlZddlmZddlmZGdddeZ ddZ dd Z d d Z d d Z ddZddZddZddZddZddZddZddZddZd d!Zd"d#Zd(d&d'ZdS))N)product)ceilcs&eZdZdfdd ZddZZS)PriorBoxNtraincsRtt|d_|d_|d_|_fddjD_d_dS)N min_sizesstepsclipcs0g|](}tjd|tjd|gqS)r)r image_size).0stepselfm/var/www/html/stable-diffusion-webui/repositories/CodeFormer/facelib/detection/retinaface/retinaface_utils.py z%PriorBox.__init__..s) superr__init__rrrr feature_mapsname)rcfgr phase __class__r rr s   zPriorBox.__init__c sg}tjD]\}j}tt|dt|dD]\}}|D]x}|jd}|jd}fdd|dfD} fdd|dfD} t| | D]\} } || | ||g7}qqFq:qt|dd} j r| j ddd | S) Nrr cs$g|]}|jjdqS)r rr )r xkrrrrrz$PriorBox.forward..?cs$g|]}|jjdqS)rr)r yrrrrr)maxmin) enumeraterrrranger torchTensorviewrclamp_)ranchorsfrijmin_sizes_kxs_kyZdense_cxZdense_cycycxoutputrrrforwards "zPriorBox.forward)Nr)__name__ __module__ __qualname__rr6 __classcell__rrrrrs rcCsBtjjt|ddddft|dddf|d}t|S)zPure Python NMS baseline.Nr#)boxesscores iou_threshold) torchvisionopsnmsr(r)list)detsthreshkeeprrr py_cpu_nms's rEcCs`t|ddddf|ddddfd|ddddf|ddddfdfdS)a Convert prior_boxes to (xmin, ymin, xmax, ymax) representation for comparison to point form ground truth data. Args: boxes: (tensor) center-size default boxes from priorbox layers. Return: boxes: (tensor) Converted xmin, ymin, xmax, ymax form of boxes. Nr r(catr;rrr point_form2s **rJcCsZt|ddddf|ddddfd|ddddf|ddddfdS)z Convert prior_boxes to (cx, cy, w, h) representation for comparison to center-size form ground truth data. Args: boxes: (tensor) point_form boxes Return: boxes: (tensor) Converted xmin, ymin, xmax, ymax form of boxes. NrFr rGrIrrr center_sizeAs *&rKcCs|d}|d}t|ddddfd||d|ddddfd||d}t|ddddfd||d|ddddfd||d}tj||dd}|dddddf|dddddfS)al We resize both tensors to [A,B,2] without new malloc: [A,2] -> [A,1,2] -> [A,B,2] [B,2] -> [1,B,2] -> [A,B,2] Then we compute the area of intersect between box_a and box_b. Args: box_a: (tensor) bounding boxes, Shape: [A,4]. box_b: (tensor) bounding boxes, Shape: [B,4]. Return: (tensor) intersection area, Shape: [A,B]. rNrFr r%)sizer(r% unsqueezeexpandr$clamp)box_abox_bABZmax_xyZmin_xyinterrrr intersectOs  LLrVcCst||}|dddf|dddf|dddf|dddfd|}|dddf|dddf|dddf|dddfd|}|||}||S)u Compute the jaccard overlap of two sets of boxes. The jaccard overlap is simply the intersection over union of two boxes. Here we operate on ground truth boxes and default boxes. E.g.: A ∩ B / A ∪ B = A ∩ B / (area(A) + area(B) - A ∩ B) Args: box_a: (tensor) Ground truth bounding boxes, Shape: [num_objects,4] box_b: (tensor) Prior boxes from priorbox layers, Shape: [num_priors,4] Return: jaccard overlap: (tensor) Shape: [box_a.size(0), box_b.size(0)] NrFrr )rVrN expand_as)rQrRrUarea_aarea_bunionrrrjaccardbs LL r\cCst|ddtjddf|ddddf}t|ddtjddf|ddddf}tj||dd||kjdd}tj|ddddf|ddddfdd}tj|ddddf|ddddfdd}||ddtjf||S)zF return iou of a and b, numpy version for data augenmentation NrFaxisr npmaximumnewaxisminimumprodall)abltrbarea_irYrZrrr matrix_iouus 00"22rkcCst|ddtjddf|ddddf}t|ddtjddf|ddddf}tj||dd||kjdd}tj|ddddf|ddddfdd}|t|ddtjfdS)zF return iof of a and b, numpy version for data augenmentation NrFr]r r_)rfrgrhrirjrYrrr matrix_iofs 00"2rlc Cs0t|t|} | jddd\} } | dddfdk} | | ddf}|jddkrfd|| <d|| <dS| jddd\}}|d|d| d|d| d|d|dt| dD]}||| |<q||}||}d|||k<t|||}||}t |||}||| <||| <||| <dS)a!Match each prior box with the ground truth box of the highest jaccard overlap, encode the bounding boxes, then return the matched indices corresponding to both confidence and location preds. Args: threshold: (float) The overlap threshold used when matching boxes. truths: (tensor) Ground truth boxes, Shape: [num_obj, 4]. priors: (tensor) Prior boxes from priorbox layers, Shape: [n_priors,4]. variances: (tensor) Variances corresponding to each prior coord, Shape: [num_priors, 4]. labels: (tensor) All the class labels for the image, Shape: [num_obj]. landms: (tensor) Ground truth landms, Shape [num_obj, 10]. loc_t: (tensor) Tensor to be filled w/ encoded location targets. conf_t: (tensor) Tensor to be filled w/ matched indices for conf preds. landm_t: (tensor) Tensor to be filled w/ encoded landm targets. idx: (int) current batch index Return: The matched indices corresponding to 1)location 2)confidence 3)landm preds. r TkeepdimNrg?rF) r\rJr$shapesqueeze_ index_fill_r'rMencode encode_landm) thresholdZtruthspriors varianceslabelslandmsloc_tZconf_tZlandm_tidxoverlapsZbest_prior_overlapZbest_prior_idxZ valid_gt_idxZbest_prior_idx_filterZbest_truth_overlapZbest_truth_idxr/matchesconflocZ matches_landmlandmrrrmatchs4        rcCs|ddddf|ddddfd|ddddf}||d|ddddf}|ddddf|ddddf|ddddf}t||d}t||gdS)aEncode the variances from the priorbox layers into the ground truth boxes we have matched (based on jaccard overlap) with the prior boxes. Args: matched: (tensor) Coords of ground truth for each prior in point-form Shape: [num_priors, 4]. priors: (tensor) Prior boxes in center-offset form Shape: [num_priors,4]. variances: (list[float]) Variances of priorboxes Return: encoded boxes (tensor), Shape: [num_priors, 4] NrFrr )r(logrH)matchedrurvg_cxcyZg_whrrrrrs @ <rrcCsFt||dddf}|dddfd|ddd}|dddfd|ddd}|dddfd|ddd}|dddfd|ddd}tj||||gdd}|ddddddf|ddddddf}||d|ddddddf}||dd}|S) aEncode the variances from the priorbox layers into the ground truth boxes we have matched (based on jaccard overlap) with the prior boxes. Args: matched: (tensor) Coords of ground truth for each prior in point-form Shape: [num_priors, 10]. priors: (tensor) Prior boxes in center-offset form Shape: [num_priors,4]. variances: (list[float]) Variances of priorboxes Return: encoded landm (tensor), Shape: [num_priors, 10] rrFNr rWdimr")r(reshaperMrNrOrH)rrurvZ priors_cxZ priors_cyZpriors_wZpriors_hrrrrrss****4&rsc Cst|ddddf|ddddf|d|ddddf|ddddft|ddddf|dfd}|ddddf|ddddfd8<|ddddf|ddddf7<|S)aDecode locations from predictions using priors to undo the encoding we did for offset regression at train time. Args: loc (tensor): location predictions for loc layers, Shape: [num_priors,4] priors (tensor): Prior boxes in center-offset form. Shape: [num_priors,4]. variances: (list[float]) Variances of priorboxes Return: decoded bounding box predictions NrFrr )r(rHexp)r~rurvr;rrrdecodes F40,rc Cs`|ddddf|ddddf|d|ddddf|ddddf|ddddf|d|ddddf|ddddf|ddddf|d|ddddf|ddddf|ddddf|d|ddddf|ddddf|ddddf|d|ddddff}tj|dd }|S) aDecode landm from predictions using priors to undo the encoding we did for offset regression at train time. Args: pre (tensor): landm predictions for loc layers, Shape: [num_priors,10] priors (tensor): Prior boxes in center-offset form. Shape: [num_priors,4]. variances: (list[float]) Variances of priorboxes Return: decoded landm predictions NrFrr# r rrG)prerurvtmprxrrr decode_landms BBBBBrc Cs|ddddddf|ddddddf|d|ddddddf|ddddddft|ddddddf|df}tj|dd}|ddddddf|ddddddfd8<|ddddddf|ddddddf7<|S)aDecode locations from predictions using priors to undo the encoding we did for offset regression at train time. Args: b_loc (tensor): location predictions for loc layers, Shape: [num_batches,num_priors,4] priors (tensor): Prior boxes in center-offset form. Shape: [1,num_priors,4]. variances: (list[float]) Variances of priorboxes Return: decoded bounding box predictions NrFrr r)r(rrH)b_locrurvr;rrrbatched_decode)s T@<8rc Cs|ddddddf|ddddddf|d|ddddddf|ddddddf|ddddddf|d|ddddddf|ddddddf|ddddddf|d|ddddddf|ddddddf|ddddddf|d|ddddddf|ddddddf|ddddddf|d|ddddddff}tj|dd}|S) aDecode landm from predictions using priors to undo the encoding we did for offset regression at train time. Args: pre (tensor): landm predictions for loc layers, Shape: [num_batches,num_priors,10] priors (tensor): Prior boxes in center-offset form. Shape: [1,num_priors,4]. variances: (list[float]) Variances of priorboxes Return: decoded landm predictions NrFrr#rrrrrG)rrurvrxrrrbatched_decode_landm@s TTTTTrcCs.|j}ttjt||ddd|S)zUtility function for computing log_sum_exp while determining This will be used to determine unaveraged confidence loss across all examples in a batch. Args: x (Variable(tensor)): conf_preds from conf layers r Trm)datar$r(rsumr)rx_maxrrr log_sum_expWs rr cCs*t|dd}|dkr*|S|dddf}|dddf}|dddf}|dddf}t||||} |d\} } | | d} |} |} |}|}|}|}d}| dkr"| d}|||<|d7}| ddkrq"| dd} tj |d| | dtj |d| | dtj |d| |dtj |d| |dtj | ||d} tj | ||d} tj |||d }tj |||d }| || ||| }|| }tj |d d}tj |d d}||}t | d| }||| |}||}| | |} q||fS) aApply non-maximum suppression at test time to avoid detecting too many overlapping bounding boxes for a given object. Args: boxes: (tensor) The location preds for the img, Shape: [num_priors,4]. scores: (tensor) The class predscores for the img, Shape:[num_priors]. overlap: (float) The overlap thresh for suppressing unnecessary boxes. top_k: (int) The Maximum number of box preds to consider. Return: The indices of the kept boxes with respect to num_priors. rNr rFrWr")outrL)r$g) r(r)rMfill_longnumelmulsortnew index_selectrP resize_as_le)r;r<overlaptop_krDx1y1x2y2areavrzxx1yy1xx2yy2whcountr.rUZ rem_areasr[ZIoUrrrr@esX     r@)r r)numpyr`r(r> itertoolsrmathrobjectrrErJrKrVr\rkrlrrrrsrrrrrr@rrrrs*     :