j/ddlmZmZmZmZmZddlZddlZddlZddlm Z m Z ddl m Z ddl mZddlmZejjd e d ee d e fd Z d+dededededef dZGddZdee d e fdZde deed efdZejjdee deeeefdeded eeeeff dZejjd eee fd!eed ee fd"Zejjd#ee dee d$eed%ed&eeed'eed e fd(ZGd)d*e jZ dS),)DictListOptionalTupleUnionN)nnTensorbox_area)_log_api_usage_once) roi_alignlevelsunmerged_resultsreturnc |d}|j|j}}tj|d|d|d|df||}t t |D]}tj||kddddd}| |d||d||d||d}| d|||}|S)Nrrr dtypedevice) rrtorchzerossizerangelenwhereviewexpandscatter)rr first_resultrrreslevelindexs HC:\Projects\Historical\venv\Lib\site-packages\torchvision/ops/poolers.py_onnx_merge_levelsr'sL#A&L & (;6E + Q**1--|/@/@/C/C\EVEVWXEYEYZbgpv   Cs+,,--== FeO,,Q/44RAqAA JJqMM U # ( ( + + U # ( ( + + U # ( ( + +    kk!U$4U$;<< Jư>k_mink_maxcanonical_scalecanonical_levelepsc(t|||||SN) LevelMapper)r,r-r.r/r0s r&initLevelMapperr4%s ue_os K KKr(c ReZdZdZ ddedededed ef d Zd eed efd Z dS)r3zDetermine which FPN level each RoI in a set of RoIs should map to based on the heuristic in the FPN paper. Args: k_min (int) k_max (int) canonical_scale (int) canonical_level (int) eps (float) r)r*r+r,r-r.r/r0cL||_||_||_||_||_dSr2)r,r-s0lvl0r0)selfr,r-r.r/r0s r&__init__zLevelMapper.__init__;s+  !# r(boxlistsrctjtjd|D}tj|jtj||jz ztj|j|j z}tj ||j |j }| tj|j z tjS)z< Args: boxlists (list[BoxList]) c,g|]}t|Sr ).0boxlists r& z(LevelMapper.__call__..Os !L!L!L(7"3"3!L!L!Lr(r)minmax)rsqrtcatfloorr8log2r7tensorr0rclampr,r-toint64)r9r;s target_lvlss r&__call__zLevelMapper.__call__Is Juy!L!L8!L!L!LMM N Nk$)ejTW.E.E"E UYU]efelHmHmHm"mnn k+4:4:NNN u{++dj8<r(r&r3r3/s   #              Jf J& J J J J J Jr(r3boxesctj|d}|j|jctjfdt |Dd}tj||gd}|S)Nr)dimc tg|]4\}}tj|ddddf|tj5S)Nr)rlayoutr)r full_likestrided)r?ibrrs r&rAz*_convert_to_roi_format..[sHuuu\`\]_`111bqb51E%-PV W W Wuuur(r)rrFrr enumerate)rW concat_boxesidsroisrrs @@r&_convert_to_roi_formatrdWs9U***L ');MFE )uuuuudmnsdtdtuuu    C 9c<(a 0 0 0D Kr(feature original_sizecX|jdd}g}t||D]\}}t|t|z }dttj|z}|||dS)Nr r)shapeziprVrrIrHroundappend)rerfrpossible_scaless1s2 approx_scalescales r& _infer_scalerrbs = D#%OdM**&&BRyy599, U5< 55::<<BBDDEEEu%%%% 1 r(features image_shapescB |stdd}d}|D].}t|d|}t|d|}/||f fd|D}tjtj|dtj }tjtj|dtj } tt|t| ||} || fS)Nzimages list should not be emptyrrc0g|]}t|Sr>)rr)r?featoriginal_input_shapes r&rAz!_setup_scales..zs$ L L L4l4!566 L L Lr(rBrr.r/) ValueErrorrDrrHrIfloat32itemr4rU) rsrtr.r/max_xmax_yriscaleslvl_minlvl_max map_levelsrxs @r& _setup_scalesrms! <:;;; E E%%E!He$$E!He$$!5> L L L L8 L L LFz%,vay FFFGGLLNNNGz%,vbzGGGHHMMOOOG  G  G '' J : r(x featmap_namescpg}|D]\}}||vr|||Sr2)itemsrl)rr x_filteredkvs r& _filter_inputrsHJ !!1     a r(r output_sizesampling_ratiormapperc6||tdt|}t|}|dkr t|d|||d|S||}t|} |djd} |dj|dj} } tj| | f|z| | } g}tt||D]\}\}}tj ||kd}||}t|||||}tj r)||| ~|| j| |<tj rt!||} | S)a Args: x_filtered (List[Tensor]): List of input tensors. boxes (List[Tensor[N, 4]]): boxes to be used to perform the pooling operation, in (x1, y1, x2, y2) format and in the image reference size, not the feature map reference. The coordinate must satisfy ``0 <= x1 < x2`` and ``0 <= y1 < y2``. output_size (Union[List[Tuple[int, int]], List[int]]): size of the output sampling_ratio (int): sampling ratio for ROIAlign scales (Optional[List[float]]): If None, scales will be automatically inferred. Default value is None. mapper (Optional[LevelMapper]): If none, mapper will be automatically inferred. Default value is None. Returns: result (Tensor) Nz$scales and mapper should not be Nonerr)r spatial_scalerr)rzrrdrrirrrrr`rjr torchvision _is_tracingrlrKr')rrWrrrr num_levelsrcrnum_rois num_channelsrrresulttracing_resultsr$per_level_featurerq idx_in_levelrois_per_levelresult_idx_in_levels r&_multiscale_roi_alignrs,~?@@@ZJ !% ( (DQ qM # ))     VE]]F4yyHa=&q)LqM'A)=6E [       FO-6s:v7N7N-O-OHH))!5{6U?33A6 l+'  #)      " $ $ H  " "#6#9#9%#@#@ A A A A$7#9#9&,#G#GF<   =#FO<< Mr(c eZdZdZeeeeedZddddee de e e e ee fde d e d e f fd Z d ee efd eedee e e fdefdZde fdZxZS)MultiScaleRoIAligna{ Multi-scale RoIAlign pooling, which is useful for detection with or without FPN. It infers the scale of the pooling via the heuristics specified in eq. 1 of the `Feature Pyramid Network paper `_. They keyword-only parameters ``canonical_scale`` and ``canonical_level`` correspond respectively to ``224`` and ``k0=4`` in eq. 1, and have the following meaning: ``canonical_level`` is the target level of the pyramid from which to pool a region of interest with ``w x h = canonical_scale x canonical_scale``. Args: featmap_names (List[str]): the names of the feature maps that will be used for the pooling. output_size (List[Tuple[int, int]] or List[int]): output size for the pooled region sampling_ratio (int): sampling ratio for ROIAlign canonical_scale (int, optional): canonical_scale for LevelMapper canonical_level (int, optional): canonical_level for LevelMapper Examples:: >>> m = torchvision.ops.MultiScaleRoIAlign(['feat1', 'feat3'], 3, 2) >>> i = OrderedDict() >>> i['feat1'] = torch.rand(1, 5, 64, 64) >>> i['feat2'] = torch.rand(1, 5, 32, 32) # this feature won't be used in the pooling >>> i['feat3'] = torch.rand(1, 5, 16, 16) >>> # create some random bounding boxes >>> boxes = torch.rand(6, 4) * 256; boxes[:, 2:] += boxes[:, :2] >>> # original image size, before computing the feature maps >>> image_sizes = [(512, 512)] >>> output = m(i, [boxes], image_sizes) >>> print(output.shape) >>> torch.Size([6, 5, 3, 3]) )rrr)r*ryrrrr.r/ctt|t|tr||f}||_||_t||_d|_ d|_ ||_ ||_ dSr2) superr:r isinstancerUrrtuplerrrr.r/)r9rrrr.r/ __class__s r&r:zMultiScaleRoIAlign.__init__ s D!!! k3 ' ' 5& 4K*, -- ..r(rrWrtrct||j}|j|j)t |||j|j\|_|_t|||j|j |j|jS)a Args: x (OrderedDict[Tensor]): feature maps for each level. They are assumed to have all the same number of channels, but they can have different sizes. boxes (List[Tensor[N, 4]]): boxes to be used to perform the pooling operation, in (x1, y1, x2, y2) format and in the image reference size, not the feature map reference. The coordinate must satisfy ``0 <= x1 < x2`` and ``0 <= y1 < y2``. image_shapes (List[Tuple[height, width]]): the sizes of each image before they have been fed to a CNN to obtain feature maps. This allows us to infer the scale factor for each one of the levels to be pooled. Returns: result (Tensor) ) rrrrrr.r/rrr)r9rrWrtrs r&forwardzMultiScaleRoIAlign.forward!s|&#1d&899 ; $/"9+8L$*>@T,, (DK%       K O    r(cP|jjd|jd|jd|jdS)Nz(featmap_names=z, output_size=z, sampling_ratio=))rrQrrr)r9s r&__repr__zMultiScaleRoIAlign.__repr__CsV~& U Ut7I U U+ U U>B>Q U U U r()rQrRrSrTrrrVr3__annotations__strrrUrr:rr rr __classcell__)rs@r&rrs/!!F"*$u+!6h{F[\\O # ///Cy/3c DI56/ /  ///////*  V   F|  5c?+       D #        r(rrP)!typingrrrrrrtorch.fxrrr torchvision.ops.boxesr utilsr rjitunusedr'rUrVr4r3rdrrfxwraprrrrModulerr>r(r&rs;55555555555555 ******'''''' vf&, LL L LL L  LLLL%J%J%J%J%J%J%J%JP$v,6&cu6l*.uS#X*?RUhk 4; #$6T#v+&tCyT&\PV P <PcP P T%[ ! P [ ! P PPPPfa a a a a a a a a a r(