jXddlZddlmZddlmZmZmZmZddlZddlm Z m Z ddl m Z ddlmZmZmZmZGddZejjd e d e d e d e fd ZGddZGddZGddZGddeZde jded dfdZde jdeeefd eefdZejj de d efdZ!de deded efd Z" d(d!e#d"ed#e d$e d%e d&eee#efd e fd'Z$dS))N) OrderedDict)DictListOptionalTuple)nnTensor) functional)complete_box_iou_lossdistance_box_iou_lossFrozenBatchNorm2dgeneralized_box_iou_losscjeZdZdZdededdfdZdeede eeeeffdZ dS) BalancedPositiveNegativeSamplerz` This class samples batches, ensuring that they contain a fixed proportion of positives batch_size_per_imagepositive_fractionreturnNc"||_||_dS)z Args: batch_size_per_image (int): number of elements to be selected per image positive_fraction (float): percentage of positive elements per batch N)rr)selfrrs TC:\Projects\Historical\venv\Lib\site-packages\torchvision/models/detection/_utils.py__init__z(BalancedPositiveNegativeSampler.__init__s %9!!2 matched_idxscHg}g}|D]}tj|dkd}tj|dkd}t|j|jz}t ||}|j|z }t ||}tj||jd|} tj||jd|} || } || } tj |tj } tj |tj }d| | <d|| <| | | |||fS)a Args: matched_idxs: list of tensors containing -1, 0 or positive values. Each tensor corresponds to a specific image. -1 values are ignored, 0 are considered as negatives and > 0 as positives. Returns: pos_idx (list[tensor]) neg_idx (list[tensor]) Returns two lists of binary masks for each image. The first list contains the positive elements that were selected, and the second list the negative example. rdeviceNdtype) torchwhereintrrminnumelrandpermr zeros_likeuint8append)rrpos_idxneg_idxmatched_idxs_per_imagepositivenegativenum_posnum_negperm1perm2pos_idx_per_imageneg_idx_per_imagepos_idx_per_image_maskneg_idx_per_image_masks r__call__z(BalancedPositiveNegativeSampler.__call__s &2 3 3 "{#9Q#>??BH{#9Q#>??BH$3d6LLMMG(..**G44G/'9G(..**G44GN8>>#3#3HOLLLXgXVEN8>>#3#3HOLLLXgXVE (  ( &+%56LTYT_%`%`%` "%*%56LTYT_%`%`%` "89 "#4 589 "#4 5 NN1 2 2 2 NN1 2 2 2 2r) __name__ __module__ __qualname____doc__r"floatrrr rr6rrrr s3S3U3t3333. T&\. eDL$v,r?wxwywwwh proposals_x1 proposals_y1 proposals_x2 proposals_y2reference_boxes_x1reference_boxes_y1reference_boxes_x2reference_boxes_y2 ex_widths ex_heightsex_ctr_xex_ctr_y gt_widths gt_heightsgt_ctr_xgt_ctr_y targets_dx targets_dy targets_dw targets_dhtargetss r encode_boxesrbJs  B B B BQQQT?,,Q//LQQQT?,,Q//LQQQT?,,Q//LQQQT?,,Q//L(A.88;;(A.88;;(A.88;;(A.88;;|+I ,JcIo-HcJ..H"%77I#&88J!C)O3H!C*$44Hx(*+i7Jx(*+j8Jei I 5666Jei Z 7888JiZZHaPPPG NrceZdZdZejdfdeeeeefdeddfdZde e d e e de e fd Z de d e de fd Z d e d e e de fdZ d e d e de fdZdS)BoxCoderz~ This class encodes and decodes a set of bounding boxes into the representation used for training the regressors. g@O@r?bbox_xform_cliprNc"||_||_dS)za Args: weights (4-element tuple) bbox_xform_clip (float) N)r?re)rr?res rrzBoxCoder.__init__s .rr=r>cd|D}tj|d}tj|d}|||}||dS)Nc,g|]}t|Sr<)len.0bs r z#BoxCoder.encode..s;;;a3q66;;;rrrD)r rH encode_singlesplit)rr=r>boxes_per_imageras rencodezBoxCoder.encodesf;;?;;;)O;;;IiQ/// $$_i@@}}_a000rc||j}|j}tj|j||}t |||}|S)z Encode a set of proposals with respect to some reference boxes Args: reference_boxes (Tensor): reference boxes proposals (Tensor): boxes to be encoded rr)rrr as_tensorr?rb)rr=r>rrr?ras rrnzBoxCoder.encode_singlesB % '/$,eFKKK 7CCr rel_codesboxesctjt|ttfdtjt|tjdd|D}tj|d}d}|D]}||z }|dkr||d}|||}|dkr||dd}|S)Nz2This function expects boxes of type list or tuple.z5This function expects rel_codes of type torch.Tensor.c8g|]}|dS)rsizerjs rrmz#BoxCoder.decode..s"444166!99444rrrD) r _assert isinstancelisttupler rHreshape decode_single)rrurvrp concat_boxesbox_sumval pred_boxess rdecodezBoxCoder.decodes utUm , , @     y%, / / C   54e444yA... "  C sNGG Q;;!))'266I'' <@@ Q;;#++GR;;Jrc||j}|dddf|dddfz }|dddf|dddfz }|dddfd|zz}|dddfd|zz}|j\}}} } |dddddf|z } |dddddf|z } |dddddf| z } |dddddf| z }tj| |j} tj||j}| |dddfz|dddfz}| |dddfz|dddfz}tj| |dddfz}tj||dddfz}tjd|j|j |z}tjd|j|j |z}||z }||z }||z}||z}tj ||||fd  d}|S) z From a set of original boxes and encoded relative box offsets, get the decoded boxes. Args: rel_codes (Tensor): encoded boxes boxes (Tensor): reference boxes. NrArrBrrCr|)maxrsrD) torr?r clampreexptensorrstackflatten)rrurvwidthsheightsctr_xctr_yrIrJrKrLdxdydwdh pred_ctr_x pred_ctr_ypred_wpred_hc_to_c_hc_to_c_w pred_boxes1 pred_boxes2 pred_boxes3 pred_boxes4rs rrzBoxCoder.decode_singlesm))qqq!tuQQQT{*1+aaad +aaad cFl*aaad cGm+BB qqq!$Q$w " $ qqq!$Q$w " $ qqq!$Q$w " $ qqq!$Q$w " $[!5 6 6 6 [!5 6 6 6&D/)E!!!T'N: '!!!T'**U111d7^; 2402D!11<:+;FMRRRU[[<:+;FMRRRU[[ 8+  8+  8+  8+ [+{K!U[\]]]eefghh r)r7r8r9r:mathrGrr;rrr rqrnrrr<rrrdrdzs* T\SWS[\gShSh / /UE5%78 /KP /  / / / /1d6l1tF|1PTU[P\1111V6 tF|*)v)f)))))))rrdcLeZdZdZd deddfdZdededefd Zd ed edefd ZdS)BoxLinearCoderz The linear box-to-box transform defined in FCOS. The transformation is parameterized by the distance from the center of (square) src box to 4 edges of the target box. Tnormalize_by_sizerNc||_dS)zq Args: normalize_by_size (bool): normalize deltas by the size of src (anchor) boxes. N)r)rrs rrzBoxLinearCoder.__init__s "3rr=r>crd|d|dzz}d|d|dzz}||dz }||dz }|d|z }|d|z }tj||||fd} |jrA|d|dz } |d|dz } tj| | | | fd} | | z } | S)aC Encode a set of proposals with respect to some reference boxes Args: reference_boxes (Tensor): reference boxes proposals (Tensor): boxes to be encoded Returns: Tensor: the encoded relative box offsets that can be used to decode the boxes. rC.r.rA.r.rBr{rD)r rr) rr=r>reference_boxes_ctr_xreference_boxes_ctr_ytarget_ltarget_ttarget_rtarget_brareference_boxes_wreference_boxes_hreference_boxes_sizes rrqzBoxLinearCoder.encodes!$v'>QWAX'X Y #v'>QWAX'X Y)9V+<<(9V+<<V$'<<V$'<<+x8XFBOOO  ! 5 / 7/&:Q Q  / 7/&:Q Q #(;"$57HJ[\bd$$$  44Grrurvc||j}d|d|dzz}d|d|dzz}|jrA|d|dz }|d|dz }tj||||fd}||z}||dz }||dz } ||dz} ||dz} tj|| | | fd} | S) a From a set of original boxes and encoded relative box offsets, get the decoded boxes. Args: rel_codes (Tensor): encoded boxes boxes (Tensor): reference boxes. Returns: Tensor: the predicted boxes with the encoded relative box offsets. .. note:: This method assumes that ``rel_codes`` and ``boxes`` have same size for 0th dimension. i.e. ``len(rel_codes) == len(boxes)``. rrCrrrrr{rD)rrrr r) rrurvrrboxes_wboxes_h list_box_sizerrrrrs rrzBoxLinearCoder.decodes$y//uV}uV}45uV}uV}45  ! 2FmeFm3GFmeFm3G!K'7G(LRTUUUM!M1Ii// i// i// i// [+{K!U[]^^^ r)T) r7r8r9r:boolrr rqrr<rrrrs 33$3$3333!f!!F!!!!F$$v$&$$$$$$rrc feZdZdZdZdZeedZddedede d d fd Z d e d e fd Z de de d e d d fdZ d S)Matchera This class assigns to each predicted "element" (e.g., a box) a ground-truth element. Each predicted element will have exactly zero or one matches; each ground-truth element may be assigned to zero or more predicted elements. Matching is based on the MxN match_quality_matrix, that characterizes how well each (ground-truth, predicted)-pair match. For example, if the elements are boxes, the matrix may contain box IoU overlap values. The matcher returns a tensor of size N containing the index of the ground-truth element m that matches to prediction n. If there is no match, a negative value is returned. r{)BELOW_LOW_THRESHOLDBETWEEN_THRESHOLDSFhigh_threshold low_thresholdallow_low_quality_matchesrNc~d|_d|_tj||kd||_||_||_dS)a Args: high_threshold (float): quality values greater than or equal to this value are candidate matches. low_threshold (float): a lower quality threshold used to stratify matches into three levels: 1) matches >= high_threshold 2) BETWEEN_THRESHOLDS matches in [low_threshold, high_threshold) 3) BELOW_LOW_THRESHOLD matches in [0, low_threshold) allow_low_quality_matches (bool): if True, produce additional matches for predictions that have only low-quality match candidates. See set_low_quality_matches_ for more details. r{rz)low_threshold should be <= high_thresholdN)rrr r}rrr)rrrrs rrzMatcher.__init__QsJ$& "$ m~57bccc,*)B&&&rmatch_quality_matrixc|dkr/|jddkrtdtd|d\}}|jr|}nd}||jk}||jk||jkz}|j||<|j ||<|jr/|tj ddn| ||||S)a Args: match_quality_matrix (Tensor[float]): an MxN tensor, containing the pairwise quality between M ground-truth elements and N predicted elements. Returns: matches (Tensor[int64]): an N tensor where N[i] is a matched gt in [0, M - 1] or a negative value indicating that prediction i could not be matched. rzENo ground-truth boxes available for one of the images during trainingzANo proposal boxes available for one of the images during trainingrDNFzall_matches should not be None) r$shape ValueErrorrrclonerrrrr r}set_low_quality_matches_)rr matched_valsmatches all_matchesbelow_low_thresholdbetween_thresholdss rr6zMatcher.__call__fs % % ' '1 , ,#)!,11 !hiii !deee!5 8 8Q 8 ? ? g  ) !--//KKK+T-??*d.@@\TXTgEgh'+'?#$&*&="#  ) Z" e%EFFFF--g{DXYYYrrrc|d\}}tj||dddfk}|d}||||<dS)ax Produce additional matches for predictions that have only low-quality matches. Specifically, for each ground-truth find the set of predictions that have maximum overlap with it (including ties); for each prediction in that set, if it is unmatched, then match it to the ground-truth with which it has the highest quality value. rrDN)rr r!)rrrrhighest_quality_foreach_gt_ gt_pred_pairs_of_highest_qualitypred_inds_to_updates rrz Matcher.set_low_quality_matches_sm)=(@(@Q(@(G(G%"A+0;7KOijkjkjkmqjqOr7r+s+s(?qA'23F'G#$$$r)F)r7r8r9r:rrr"__annotations__r;rrr r6rr<rrrr:s   #!O CCuCUC_cCptCCCC*&V&&&&&PHHVHciHnrHHHHHHrrc<eZdZdeddffd Zdedeffd ZxZS) SSDMatcher thresholdrNcPt||ddS)NF)r)superr)rr __class__s rrzSSDMatcher.__init__s( IOOOOOrrct|}|d\}}tj|dtj|j||<|S)NrrDrrs)rr6rr arangerzint64r)rrrrhighest_quality_pred_foreach_gtrs rr6zSSDMatcher.__call__sx''""#788.B-E-E!-E-L-L* *38< + 0 0 3 35;OnOu4 4 4 /0r)r7r8r9r;rr r6 __classcell__)rs@rrrsP%PDPPPPPP V           rrmodelepscl|D]}t|tr||_dS)a This method overwrites the default eps values of all the FrozenBatchNorm2d layers of the model with the provided value. This is necessary to address the BC-breaking change introduced by the bug-fix at pytorch/vision#2933. The overwrite is applied only when the pretrained weights are loaded to maintain compatibility with previous versions. Args: model (nn.Module): The model on which we perform the overwrite. eps (float): The new value of eps. N)modulesr~r r)rrmodules r overwrite_epsrs@--// f/ 0 0 FJrrzc|j}|tj5t |j}tjdd|d|df|}||}t|tj rtd|fg}d| D}dddn #1swxYwY|r| |S)a This method retrieves the number of output channels of a specific model. Args: model (nn.Module): The model for which we estimate the out_channels. It should return a single Tensor or an OrderedDict[Tensor]. size (Tuple[int, int]): The size (wxh) of the input. Returns: out_channels (List[int]): A list of the output channels of the model. rrBrr0c8g|]}|dS)rry)rkxs rrmz)retrieve_out_channels..s"===aq ===rN) trainingevalr no_gradnext parametersrzerosr~r rvaluestrain)rrz in_trainingrtmp_imgfeatures out_channelss rretrieve_out_channelsrs).K JJLLL >>e&&(())0+q!T!Wd1g6vFFF5>> h - - 6"S(O#455H==8??+<+<=== >>>>>>>>>>>>>>> sB"CC!$C!vc|SNr<)rs r_fake_cast_onnxrs Hrinput orig_kvalaxisctjs#t|||Stj||d}tjtjtj|g|j |fd}t|S)ay ONNX spec requires the k-value to be less than or equal to the number of inputs along provided dim. Certain models use the number of elements along a particular axis instead of K if K exceeds the number of elements along that axis. Previously, python's min() function was used to determine whether to use the provided k-value or the specified dim axis value. However, in cases where the model is being exported in tracing mode, python min() is static causing the model to be traced incorrectly and eventually fail at the topk node. In order to avoid this situation, in tracing mode, torch.min() is used instead. Args: input (Tensor): The original input tensor. orig_kval (int): The provided k-value. axis(int): Axis along which we retrieve the input size. Returns: min_kval (int): Appropriately selected k-value. rr) r jit is_tracingr#rz_shape_as_tensorrFrHrrr)rrr axis_dim_valmin_kvals r _topk_minrs& 9   ! !09ejj..///)%006@@CCLyEL)LDV$W$W$WYe#fhijjkkH 8 $ $$rtype box_coderanchors_per_imagematched_gt_boxes_per_imagebbox_regression_per_imagecnfctj|dvd||dkr-|||}tj||dS|dkr>|||}| d|vr|dnd}tj||d| S|||}| d |vr|d nd } |d krt||d| S|dkrt||d| St||d| S)N)l1 smooth_l1cioudiougiouzUnsupported loss: r sum) reductionr betag?)rrrgHz>r )rrr) r r}rnFl1_losssmooth_l1_lossrr r r) rrrrrr target_regressionrbbox_per_imagers r _box_lossrsK M$EEGb\`GbGbccc t||%334NPabby24EQVWWWW   %334NPabb!o&C--s6{{S 9;LX]dhiiii"001JL]^^O c%jj$ 6>>(9S_djmnnn n 6>>(9S_djmnnn n'8R^cilmmmmrr)%r collectionsrtypingrrrrr rr torch.nnr rtorchvision.opsr r r rrr_script_if_tracingrbrdrrrModuler;rr"runusedrrstrrr<rrr!s ######............ $$$$$$uuuuuuuuuuuu< < < < < < < < ~,&,V,f,QW,,,,^ffffffffRTTTTTTTTngHgHgHgHgHgHgHgHT         4$%S/d3i< v #    %V%%3%3%%%%@'+ nn nnn!' n & n $sEz" # n nnnnnnr