U (d-V @sXddlZddlmZddlmZmZmZmZddlZddlm Z m Z ddl m Z ddlmZmZmZmZGdddZejje e e e d d d ZGd d d ZGdddZGdddZGdddeZe jeddddZe jeeefeedddZejj e edddZ!e eeedddZ"d#e#ee e e eee#efe d d!d"Z$dS)$N) OrderedDict)DictListOptionalTuple)Tensornn) functional)FrozenBatchNorm2dcomplete_box_iou_lossdistance_box_iou_lossgeneralized_box_iou_lossc@sFeZdZdZeeddddZeee eeeefdddZ dS) BalancedPositiveNegativeSamplerz` This class samples batches, ensuring that they contain a fixed proportion of positives N)batch_size_per_imagepositive_fractionreturncCs||_||_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)selfrrrG/tmp/pip-unpacked-wheel-vx7f76es/torchvision/models/detection/_utils.py__init__sz(BalancedPositiveNegativeSampler.__init__) matched_idxsrcCsg}g}|D]}t|dkd}t|dkd}t|j|j}t||}|j|}t||}tj||jdd|} tj||jdd|} || } || } tj |tj d} tj |tj d}d| | <d|| <| | | |q ||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) torchwhereintrrminnumelZrandpermrZ zeros_likeZuint8append)rrZpos_idxZneg_idxZmatched_idxs_per_imageZpositivenegativenum_posZnum_negZperm1Zperm2Zpos_idx_per_imageZneg_idx_per_imageZpos_idx_per_image_maskZneg_idx_per_image_maskrrr__call__s(   z(BalancedPositiveNegativeSampler.__call__) __name__ __module__ __qualname____doc__rfloatrrrrr$rrrrr s r)reference_boxes proposalsweightsrcCs~|d}|d}|d}|d}|dddfd}|dddfd}|dddfd} |dddfd} |dddfd} |dddfd} |dddfd} |dddfd}| |}| |}|d|}|d|}| | }|| }| d|}| d|}||||}||||}|t||}|t||}tj||||fdd}|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 weights (Tensor[4]): the weights for ``(x, y, w, h)`` rrN?Zdim) unsqueezerlogcat)r*r+r,wxwywwwhZ proposals_x1Z proposals_y1Z proposals_x2Z proposals_y2Zreference_boxes_x1Zreference_boxes_y1Zreference_boxes_x2Zreference_boxes_y2Z ex_widthsZ ex_heightsZex_ctr_xZex_ctr_yZ gt_widthsZ gt_heightsZgt_ctr_xZgt_ctr_yZ targets_dxZ targets_dyZ targets_dwZ targets_dhtargetsrrr encode_boxesJs4     r9c@seZdZdZedfeeeeefeddddZe e e e e e ddd Z e e e dd d Z e e e e d d dZ e e e d ddZdS)BoxCoderz~ This class encodes and decodes a set of bounding boxes into the representation used for training the regressors. g@O@N)r,bbox_xform_cliprcCs||_||_dS)za Args: weights (4-element tuple) bbox_xform_clip (float) N)r,r;)rr,r;rrrrszBoxCoder.__init__r*r+rcCsBdd|D}tj|dd}tj|dd}|||}||dS)NcSsg|] }t|qSr)len.0brrr sz#BoxCoder.encode..rr0)rr3 encode_singlesplit)rr*r+boxes_per_imager8rrrencodes  zBoxCoder.encodecCs.|j}|j}tj|j||d}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)rrrZ as_tensorr,r9)rr*r+rrr,r8rrrrBs  zBoxCoder.encode_single rel_codesboxesrcCstt|ttfdtt|tjddd|D}tj|dd}d}|D] }||7}qN|dkrp||d}|||}|dkr||dd}|S) Nz2This function expects boxes of type list or tuple.z5This function expects rel_codes of type torch.Tensor.cSsg|]}|dqS)rsizer>rrrrAsz#BoxCoder.decode..rr0) r_assert isinstancelisttuplerr3Zreshape decode_single)rrHrIrDZ concat_boxesZbox_sumval pred_boxesrrrdecodes&     zBoxCoder.decodecCs||j}|dddf|dddf}|dddf|dddf}|dddfd|}|dddfd|}|j\}}} } |dddddf|} |dddddf|} |dddddf| } |dddddf| }tj| |jd} tj||jd}| |dddf|dddf}| |dddf|dddf}t| |dddf}t||dddf}tjd|j|jd |}tjd|j|jd |}||}||}||}||}tj ||||fdd  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. Nr-rr.rr/rM)maxrFr0) torr,rclampr;exptensorrstackflatten)rrHrIZwidthsZheightsctr_xctr_yr4r5r6r7ZdxZdyZdwZdhZ pred_ctr_xZ pred_ctr_yZpred_wZpred_hZc_to_c_hZc_to_c_w pred_boxes1 pred_boxes2 pred_boxes3 pred_boxes4rTrrrrRs0   $$zBoxCoder.decode_single)r%r&r'r(mathr2rr)rrrrErBrUrRrrrrr:zs r:c@sFeZdZdZd eddddZeeeddd Zeeed d d ZdS)BoxLinearCoderz The linear box-to-box transform defined in FCOS. The transformation is parameterized by the distance from the center of (square) src box to 4 edges of the target box. TN)normalize_by_sizercCs ||_dS)zq Args: normalize_by_size (bool): normalize deltas by the size of src (anchor) boxes. N)re)rrerrrrszBoxLinearCoder.__init__r<c Csd|dddf|dddf}d|dddf|dddf}||dddf}||dddf}|dddf|}|dddf|}tj||||fdd} |jr|dddf|dddf} |dddf|dddf} tj| | | | fdd} | | } | S)aB Encode a set of proposals with respect to some reference boxes Args: reference_boxes (Tensor): reference boxes proposals (Tensor): boxes to be encoded Returns: Tensor: the encoded relative box offsets that can be used to decode the boxes. r/Nrr-rr.r0)rr[re) rr*r+Zreference_boxes_ctr_xZreference_boxes_ctr_yZtarget_lZtarget_tZtarget_rZtarget_br8Zreference_boxes_wZreference_boxes_hZreference_boxes_sizerrrrBs $$   zBoxLinearCoder.encode_singlerGc Cs"||j}d|dddf|dddf}d|dddf|dddf}|jr|dddf|dddf}|dddf|dddf}tj||||fdd}||}||dddf}||dddf} ||dddf} ||dddf} tj|| | | fdd} | 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. r/Nrr-rr.r0)rWrrerr[) rrHrIr]r^Zboxes_wZboxes_hZ boxes_sizer_r`rarbrTrrrrRs $$  zBoxLinearCoder.decode_single)T) r%r&r'r(boolrrrBrRrrrrrds!rdc@s\eZdZdZdZdZeedZdeee dddd Z e e d d d Z e e e dd ddZ dS)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. rL)BELOW_LOW_THRESHOLDBETWEEN_THRESHOLDSFN)high_threshold low_thresholdallow_low_quality_matchesrcCs2d|_d|_t||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. rLrhz)low_threshold should be <= high_thresholdN)rirjrrNrkrlrm)rrkrlrmrrrrGs zMatcher.__init__match_quality_matrixrcCs|dkr,|jddkr$tdntd|jdd\}}|jrL|}nd}||jk}||jk||jk@}|j||<|j ||<|jr|dkrt ddn| ||||S)a Args: match_quality_matrix (Tensor[float]): an MxN tensor, containing the pairwise quality between M ground-truth elements and N predicted elements. Returns: matches (Tensor[int64]): an N tensor where N[i] is a matched gt in [0, M - 1] or a negative value indicating that prediction i could not be matched. rzENo ground-truth boxes available for one of the images during trainingzANo proposal boxes available for one of the images during trainingr0NFzall_matches should not be None) r shape ValueErrorrVrmclonerlrkrirjrrNset_low_quality_matches_)rroZ matched_valsmatches all_matchesZbelow_low_thresholdZbetween_thresholdsrrrr$\s"      zMatcher.__call__)rtrurorcCsB|jdd\}}t||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. rr0N)rVrr)rrtruroZhighest_quality_foreach_gt_Z gt_pred_pairs_of_highest_qualityZpred_inds_to_updaterrrrss z Matcher.set_low_quality_matches_)F)r%r&r'r(rirjr__annotations__r)rfrrr$rsrrrrrg0s(rgcs8eZdZeddfdd Zeedfdd ZZS) SSDMatcherN) thresholdrcstj||dddS)NF)rm)superr)rry __class__rrrszSSDMatcher.__init__rncs>t|}|jdd\}}tj|dtj|jd||<|S)Nrr0rrF)rzr$rVrZarangerKZint64r)rrortrvZhighest_quality_pred_foreach_gtr{rrr$s  zSSDMatcher.__call__)r%r&r'r)rrr$ __classcell__rrr{rrxsrx)modelepsrcCs"|D]}t|tr||_qdS)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)modulesrOr r)r~rmodulerrr overwrite_epss  r)r~rKrc Cs|j}|tft|j}tjdd|d|df|d}||}t|tj rft d|fg}dd| D}W5QRX|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. rr.rr0cSsg|]}|dqS)rrJ)r?xrrrrAsz)retrieve_out_channels..) ZtrainingevalrZno_gradnext parametersrzerosrOrrvaluesZtrain)r~rKZ in_trainingrZtmp_imgfeaturesZ out_channelsrrrretrieve_out_channelss   r)vrcCs|S)Nr)rrrr_fake_cast_onnxsr)input orig_kvalaxisrcCsZtjst|||St||d}tttj|g|j d|fd}t |S)aw 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 orignal input tensor. orig_kval (int): The provided k-value. axis(int): Axis along which we retreive the input size. Returns: min_kval (int): Appropriately selected k-value. rr) rjit is_tracingrrKZ_shape_as_tensorr1r3rZrr)rrrZ axis_dim_valZmin_kvalrrr _topk_mins  $r)type box_coderanchors_per_imagematched_gt_boxes_per_imagebbox_regression_per_imagecnfrc Cst|dkd||dkr:|||}tj||ddS|dkr||||}|dk rfd|krf|dnd}tj||d|d S|||}|dk rd |kr|d nd } |d krt||d| d S|dkrt||d| d St ||d| d SdS)N)l1 smooth_l1cioudiouZgiouzUnsupported loss: rsum) reductionrbetag?)rrrgHz>r)rrr) rrNrBFZl1_lossZsmooth_l1_lossrRr r r ) rrrrrrZtarget_regressionrZbbox_per_imagerrrr _box_losss   r)N)%rc collectionsrtypingrrrrrrrZtorch.nnr rZtorchvision.opsr r r r rrZ_script_if_tracingr9r:rdrgrxModuler)rrrZunusedrrstrrrrrrs8  ?/iMr