^jV@dZddlmZddlZddlZddlmcmZ ddl m Z ddlmZddl m Z ddlmZddlmZmZmZmZdd lmZeZd Zd ZGd d ej4ZddZy)zFModules to compute the matching cost and solve the corresponding LSAP.) annotationsN)linear_sum_assignment)nn)compute_keypoint_matching_cost) point_sample)batch_dice_lossbatch_sigmoid_ce_lossbox_cxcywh_to_xyxygeneralized_box_iou) get_loggerg?g@ceZdZdZ d dfd ZeddZej d d dZ xZ S) HungarianMatchera,This class computes an assignment between the targets and the predictions of the network For efficiency reasons, the targets don't include the no_object. Because of this, in general, there are more predictions than targets. In this case, we do a 1-to-1 matching of the best predictions, while the others are un-matched (and thus treated as non-objects). Note: The focal loss exponent ``gamma`` is fixed at ``_FOCAL_LOSS_GAMMA`` (2.0) and is not configurable. Only ``focal_alpha`` can be adjusted at construction time. c t|||_||_||_|dk7s|dk7s |dk7sJd||_||_||_| |_| xsg|_ | |_ | |_ | |_ ||_ d|_y)aCreates the matcher. Args: cost_class: Relative weight of the classification error in the matching cost. cost_bbox: Relative weight of the L1 error of the bounding box coordinates. cost_giou: Relative weight of the GIoU loss of the bounding box. focal_alpha: Alpha parameter for focal loss used in the classification cost. use_pos_only: Reserved for future use; currently has no effect. use_position_modulated_cost: Reserved for future use; currently has no effect. mask_point_sample_ratio: Downsampling ratio for mask point sampling. cost_mask_ce: Relative weight of the binary cross-entropy mask cost. cost_mask_dice: Relative weight of the Dice mask cost. rzall costs can't be 0FN)super__init__ cost_class cost_bbox cost_giou focal_alphamask_point_sample_ratio cost_mask_cecost_mask_dicenum_keypoints_per_classkeypoint_l1_loss_coefkeypoint_findable_loss_coefkeypoint_visible_loss_coefkeypoint_nll_loss_coef_warned_non_finite_costs)selfrrrr use_pos_onlyuse_position_modulated_costrrrrrrrr __class__s `/var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/rfdetr/models/matcher.pyrzHungarianMatcher.__init__2s< $""Q)q.INZDZZB&'>$(,'>'D"$%:"+F(*D'&<#(-%c4tj|}|jr|Stj|j}|j r||}|j }||jj ztz}tj|s|j|j }n=tj||j }n|j|j }|j}|||<|S)aReplace non-finite cost entries with a large finite sentinel. >>> HungarianMatcher._sanitize_cost_matrix( ... torch.tensor([[1.0, float("nan")], [float("inf"), -2.0]]) ... ).tolist() [[1.0, 4.0], [4.0, -2.0]] Args: cost_matrix: Cost matrix to sanitize before Hungarian assignment. Returns: Cost matrix with all non-finite entries replaced by a finite sentinel that is no smaller than any valid entry. )max) torchisfiniteallfinfodtypeanyr&abs_SANITIZED_COST_MARGIN new_tensorclampclone) cost_matrix finite_mask dtype_info finite_costsmax_costreplacement_costsanitized_cost_matrixs r#_sanitize_cost_matrixz&HungarianMatcher._sanitize_cost_matrix`s nn[1 ??  [[!2!23 ?? &{3L#'')H (,*:*:*<*@*@*BBE[[ >>"23#.#9#9*..#I #(;;/?Z^^#T  +55jnnE  + 1 1 3.>{l+$$r$c  |djdd\}}|djdd}|j}|djdd}tj|D cgc]} | d c} } tj|D cgc]} | d c} } d} d |dv} d |vxrd |dv}|r(tj|D cgc]} | d  c} d } t t |t | }| }|j}t}d|z ||zztj|  z}|d|z |zztj| z}|dd| f|dd| fz }tj|| d }| r/tj|D cgc]} | d  c} }t|dtjr|djdd}|jd|jdz|jz}tjd|d|j }t#|j%d|j'|jddddj)d}n|dd}|dd}|dd}|jd|jdz|jz}tjd|d|j }t#||j'|jdddd}tj*d|||z}|jdd}|j-|j.}t#|j%d|j'|jdddddj)d}t1||} t3||}!|r| }| dddf| dddfz}"t5|d | | |"|j6\}#}$}%}&|#jdd}#|$jdd}$|%jdd}%|&jdd}&|j8|z|j:|zz|j<|zz}'| r"|'|j> zz|j@!zz}'|rB|'|jB#zz|jD$zz|jF%zz|jH&zz}'|'jK||djMjO}'tjP|'}(|(jSs9|jTstVjYdd|_*|j[|'}'|D cgc]} t]| d})} g}*||z}+|'j_|+d },ta|D]}-|,|-}.tc|.j_|)dD/0cgc]\}/}0te|0|/}1}/}0|-dk(r|1}*Itg|*|1D23cgc]E\}2}3tijj|2d|3d|+|-zzgtijj|2d|3dgfG}*}2}3|*D/4cgc]O\}/}4tjl|/tjntjl|4tjnfQc}4}/Scc} wcc} wcc} wcc} wcc} wcc}0}/wcc}3}2wcc}4}/w)aPerforms the matching Args: outputs: Dict containing at least these entries: "pred_logits": Tensor of dim [batch_size, num_queries, num_classes] with the classification logits "pred_boxes": Tensor of dim [batch_size, num_queries, 4] with the predicted box coordinates targets: List of targets (len(targets) = batch_size), where each target is a dict containing: "labels": Tensor of dim [num_target_boxes] (where num_target_boxes is the number of ground-truth objects in the target) containing the class labels "boxes": Tensor of dim [num_target_boxes, 4] containing the target box coordinates "masks": Tensor of dim [num_target_boxes, H, W] containing the target mask coordinates group_detr: Number of groups used for matching. Returns: A list of size batch_size, containing tuples of (index_i, index_j) where: - index_i is the indices of the selected predictions (in order) - index_j is the indices of the corresponding selected targets (in order) For each batch element, it holds: len(index_i) = len(index_j) = min(num_queries, num_target_boxes) pred_logitsNr pred_boxeslabelsboxesmaskspred_keypoints keypoints)dim)p pred_masks)deviceF) align_cornersspatial_featuresquery_featuresbiasz bcp,bnc->bnpnearest)rJmode)all_pred_keypointstarget_keypointstarget_classes target_areasrzsNon-finite values detected in matcher cost matrix; replacing with finite sentinel. Check for numerical instability.T)r+)8shapeflattensigmoidr'catr r r_FOCAL_LOSS_GAMMAF logsigmoidcdist isinstanceTensorrrandrIr unsqueezerepeatsqueezeeinsumtor+r rrrrrrrrrrrrviewfloatcpur(r)rloggerwarningr9lensplitrange enumeraterzipnp concatenate as_tensorint64)5routputstargets group_detrbs num_queriesflat_pred_logitsout_probout_bboxvtgt_idstgt_bbox tgt_keypoints masks_presentkeypoints_presentgiouralphagammaneg_cost_classpos_cost_classrr tgt_masks out_masks num_points point_coordspred_masks_logitsrKrLrMtgt_masks_flatrrrTcost_l1 cost_findable cost_visiblecost_nllr2r3sizesindices g_num_queriescost_matrix_listg_igrouped_cost_matrixic indices_gindice1indice2js5 r#forwardzHungarianMatcher.forwards6"-066r:K#=199!Q?#++-<(00A6))':QQx[:;99':Qaj:; 71:- ,7UK7ST:]^#AwJ/.G2LL KK(a8  w"?!1W:"?@I',/>#L199!Q? &__R09??23FF$JfJff $zz!Z9CSCST $0''*L,?,? PQ@RTUWX,Yin%'!*"$+<#89K#L !(!67G!H|,V4-33B7:J:P:PQS:TTX\XtXtt $zz!ZCSCZCZ[ $0$l&9&9:J:P:PQR:SUVXY&Zjo%!%*LLARTb$cfj$j!$5$=$=a$C!! %6%<%<=I)##A&##IOOA$61=#  gaj 11BNSL-->ON !:#AqD>HQTN:L=[#*+;# :G]L(ooa+G)11!Q7M'//15L''1-Hnny04??Z3OORVR`R`clRll %(9(9L(HH4K^K^aoKooK ,,w6722]BC11L@A--8 9    Rb 1 7 7 9 = = ?  nn[1  007 15-44[AK*12QQwZ22#z1 &,,],B$ C"23"7 AJK^KdKdejlnKoApqA.qt4qIqax#-0,C  ) GAJQTAT4T'UV GAJ'?@ lsscgcdfg%++6QVQ\Q\8]^ss;: 'H*#@d3 rts2# Z Z Z$ Z)Z.<Z3.A Z9AZ?)r=r=r=g?FFr=r=Nrrr)rrfrrfrrfrrfr boolr!rrintrrfrrfrzlist[int] | Nonerrfrrfrrfrrf)r2 torch.Tensorreturnr)r=)rsdictrtlistrurrz'list[tuple[torch.Tensor, torch.Tensor]]) __name__ __module__ __qualname____doc__r staticmethodr9r'no_gradr __classcell__)r"s@r#rr&s, !",1') !48'*-0,/(+,.,.,. ,.  ,.  ,.&*,."%,.,.,."2,. %,.&+,.%*,.!&,.\%%%%NU]]_  atatat at 1 atatr$rc d|j|j|j|jt |dgt |ddt |ddt |ddt |ddd }|j r/t d i||j|j|jdSt d i|S) aoBuild a HungarianMatcher from a training argument namespace. Args: args: Namespace supplying ``focal_alpha``, ``set_cost_class``, ``set_cost_bbox``, ``set_cost_giou``, ``segmentation_head``, and optional keypoint cost coefficients (``keypoint_l1_loss_coef``, ``keypoint_findable_loss_coef``, ``keypoint_visible_loss_coef``, ``keypoint_nll_loss_coef``). When ``segmentation_head`` is truthy, also requires ``mask_ce_loss_coef``, ``mask_dice_loss_coef``, and ``mask_point_sample_ratio``. Returns: Configured HungarianMatcher instance. rrrrrr) rrrrrrrrr)rrr) set_cost_class set_cost_bbox set_cost_giourgetattrsegmentation_headrmask_ce_loss_coefmask_dice_loss_coefr)args common_kwargss r# build_matcherr-s ))''''''#*41JB#O!(/F!L'.t5RTW'X&-d4PRU&V")$0H#"N M   //33$($@$@    0-00r$)rr)r __future__rnumpyror'torch.nn.functionalr functionalrZscipy.optimizerrfdetr.models.heads.keypointsr rfdetr.models.heads.segmentationrrfdetr.utilities.box_opsrr r r rfdetr.utilities.loggerr rhr.rYModulerrrr$r#rs\$M" 0H9tt. DtryyDtN"1r$