NgHC XdZ ddlmZmZmZddlZddlZddlm Z ddl m Z m Z ddl mZmZmZmZmZddlmZdZd Zdd efd Zd ejdejfdZ ddeejdeejdedefdZdeeefdefdZGdde jZGdde Z!dS)z RetinaNet / EfficientDet Anchor Gen Adapted for PyTorch from Tensorflow impl at https://github.com/google/automl/blob/6f6694cec1a48cdb33d5d1551a2d5db8ad227798/efficientdet/anchors.py Hacked together by Ross Wightman, original copyright below )OptionalTupleSequenceN) batched_nmsremove_small_boxes) ArgMaxMatcherFasterRcnnBoxCoderBoxList IouSimilarityTargetAssigner)batched_soft_nmsggjF output_xyxyc:|dddf|dddfzdz }|dddf|dddfzdz }|dddf|dddfz }|dddf|dddfz }|d\}}} } tj| |z} tj| |z} ||z|z} ||z|z}| | dz z }|| dz z }| | dz z}|| dz z}|rtj||||gd}ntj||||gd}|S)aTransforms relative regression coordinates to absolute positions. Network predictions are normalized and relative to a given anchor; this reverses the transformation and outputs absolute coordinates for the input image. Args: rel_codes: box regression targets. anchors: anchors on all feature levels. Returns: outputs: bounding boxes. Nrr dim@)unbindtorchexpstack) rel_codesanchorsr ycenter_a xcenter_ahawatytxthtwwhycenterxcenteryminxminymaxxmaxouts J/var/www/html/ai-engine/env/lib/python3.11/site-packages/effdet/anchors.pydecode_box_outputsr.-suAA.!3IAA.!3I AA &B AA &B%%!%,,NBB " A " A2g !G2g !G QV D QV D QV D QV D;k4tT2:::k4tT2::: Jboxessizec|d}tj||gd}||}|S)Nr)minr)clamprcatr3)r0r1s r-clip_boxes_xyxyr6RsA KKAK  E 9dD\q ) ) )D IIdOOE Lr/d img_scaleimg_sizemax_det_per_imagesoft_nmsc 6|jddksJ|jddksJ|jddksJ||ddf}t||d} ||t| ||z } |d} |rt | | |ddd \} } | | | <nt| | |d } | d|} | | } | | df} || dfdz}|| |z} t| } tj | | |gd }| |kr4>A+=>> F FE$&N5 ! %do 6 6 F F "0FFF"5)00$Q-*U:e3DQ3G-H-HH$Q-*U:e3DQ3G-H-HHJ%do(>(>>+EDN,BCEFFFFF Fr/c 6g}|jD]\}}g}|D]u}|\}}}} | |dzd|zz} | |dzd|zz} t|tr|\} } nt j|} d| z } | | zdz }| | zdz }t j|ddz |jd|d}t j|ddz |jd|d}t j||\}}| d}| d}t j ||z ||z ||z||zf}t j |dd}| t j |dwt j|d}| | ddgt j |}tj|}|S) z#Generates multi-scale anchor boxes.r rrg?rr=)axisr>)rlitemsrhrnpsqrtarangerWmeshgridreshapevstackswapaxesr[ expand_dims concatenater from_numpyrG)ro boxes_allr^configs boxes_levelrlstride octave_scalerzrpbase_anchor_size_xbase_anchor_size_yaspect_xaspect_yanchor_size_x_2anchor_size_y_2xyxvyvr0rNs r-rnzAnchors._generate_boxess. +++-- ; ;JAwK! B B=C: fl%1F1I%=\@Q%Q"%1F1I%=\@Q%Q"fh//.)/&Hhh!wvH"X~H"4x"?#"E"4x"?#"EIfQi!mT_Q-?KKIfQi!mT_Q-?KKQ**BZZ^^ZZ^^ 2#7o9M#%#7o9M#OPP E1a00""2>%a#@#@#@AAAA.1===K   [00"a99 : : : :y++ ' 55;;== r/c:|jt|jzSrs)rfrJrg)ros r-get_anchors_per_locationz Anchors.get_anchors_per_locationsT%7!8!888r/) __name__ __module__ __qualname____doc__rintrd classmethodrurkrnr __classcell__)rqs@r-raras""%>bghkmphpbq%>%>%>%>%>%>N44[4   """H9999999r/rac4eZdZdZd dedefdZd dZd dZd S) AnchorLabelerz)Labeler for multiscale anchor boxes. rB num_classesmatch_thresholdct}t||dd}t}t||||_||_||_||_i|_dS)afConstructs anchor labeler to assign labels to anchors. Args: anchors: an instance of class Anchors. num_classes: integer number representing number of classes in the dataset. match_threshold: float number between 0 and 1 representing the threshold to assign positive labels for anchors. T)unmatched_thresholdnegatives_lower_than_unmatchedforce_match_for_each_rowN) r rr r target_assignerrrr indices_cache)rorrrsimilarity_calcmatcher box_coders r-rdzAnchorLabeler.__init__!ss(//  /+/%) +++ '(( -ow RR .&r/Tc>g}g}|r|dk}||}||}|jt|jjt||\}}} |dz } d} t |jj|jjdzD]} |jj | } | d| dz|j z} | || | | z | d| ddg| || | | z | d| ddg| | z } | j dk}|||fS)aHLabels anchors with ground truth inputs. Args: gt_boxes: A float tensor with shape [N, 4] representing groundtruth boxes. For each row, it stores [y0, x0, y1, x1] for four corners of a box. gt_classes: A integer tensor with shape [N, 1] representing groundtruth classes. filter_valid: Filter out any boxes w/ gt class <= -1 before assigning Returns: cls_targets_dict: ordered dictionary with keys [min_level, min_level+1, ..., max_level]. The values are tensor with shape [height_l, width_l, num_anchors]. The height_l and width_l represent the dimension of class logits at l-th level. box_targets_dict: ordered dictionary with keys [min_level, min_level+1, ..., max_level]. The values are tensor with shape [height_l, width_l, num_anchors * 4]. The height_l and width_l represent the dimension of bounding box regression output at l-th level. num_positives: scalar tensor storing number of positives in an image. r=r r)rassignr rr0longrZrerXr]rr[r match_resultsrGsum)rogt_boxes gt_classes filter_validcls_targets_outbox_targets_out valid_idx cls_targets box_targetsmatchescountrxr\steps num_positivess r- label_anchorszAnchorLabeler.label_anchors:s,  /"RI *H#I.J,0,@,G,G DL& ' '):):J-H-H) ['#Q,,..  @4<14<3IA3MNN  E /6IaL9Q</$,2W2W2Y2YYE  " ";uUU]/B#C#K#KYWX\[def[gikLl#m#m n n n  " ";uUU]/B#C#K#KYWX\[def[gikLl#m#m n n n UNEE .3::<<@@BB >>r/c t|}|t|ksJ|jj|jjz dz}dt |D}dt |D}g}t |jj} t |D]<} | |dz k} |r6|| dk} t || | } || | }nt || } || }|j| | |\}}}|dz } d}t |jj|jjdzD] }||jjz }|jj |}|d|dz|j z}|| ||||z |d|ddg|| ||||z |d|ddg||z }| r:tj||||<tj||||<"| |jdk| rtj|}>|||fS)Nr cg|]}gSr.0r^s r- z5AnchorLabeler.batch_label_anchors..p999!2999r/cg|]}gSrrrs r-rz5AnchorLabeler.batch_label_anchors..qrr/r=r)rJrrXrerZr r0rrrr]rr[rrrrrGr)rorrr batch_size num_levelsrrnum_positives_outanchor_box_listi last_sampler gt_box_list gt_class_irrrrrx level_idxr\rs r-batch_label_anchorsz!AnchorLabeler.batch_label_anchorsls]] S__,,,,\+dl.DDqH 99uZ'8'899999uZ'8'8999!$,"455z""! C! CAzA~-K +&qMB. %hqk)&<== ']95 %hqk22 '] 040D0K0KO]hjt0u0u -Kg'?0022K DEt|5t|7MPQ7QRR Y Y!DL$::  L3E: !! y|3dl6[6[6]6]] *11eem 34<rs* -,,,,,,,,, AAAAAAAAmmmmmmmmmmmmmm&&&&&&""""""J5<u|8=IIEL)I5=el5KII15IIIIX uS#X 3     e9e9e9e9e9bie9e9e9P{C{C{C{C{CF{C{C{C{C{Cr/