:j}.ddlmZmZddlZddlZddlZddlmZmZddlm Z ddl m Z ddl m Z ejjd ed eed efd Z d(d ededededef dZGddZdeed efdZdedeed efdZej.j0deedeeeefdeded eeeeff dZej.j0deeefdeed eefdZej.j0d eedeed!eed"ed#eeed$eed efd%ZGd&d'ej>Z y)))OptionalUnionN)nnTensor)box_area)_log_api_usage_once) roi_alignlevelsunmerged_resultsreturnc Z|d}|j|j}}tj|j d|j d|j d|j df||}t t |D]}tj||k(djdddd}|j|j d||j d||j d||j d}|jd|||}|S)Nrr rdtypedevice) rrtorchzerossizerangelenwhereviewexpandscatter)r r first_resultrrreslevelindexs l/home/mcse/projects/srt_converter/srt-converter-venv/lib/python3.12/site-packages/torchvision/ops/poolers.py_onnx_merge_levelsr#s"#A&L && (;(;6E ++ Q**1-|/@/@/C\EVEVWXEYZbgpv Cs+,-= FeO,Q/44RAqA JJqM U # ( ( + U # ( ( + U # ( ( +   kk!U$4U$;<= Jk_mink_maxcanonical_scalecanonical_levelepsc t|||||SN) LevelMapper)r%r&r'r(r)s r"initLevelMapperr-%s ue_os KKr$c JeZdZdZ d dededededef dZdeed efd Z y ) r,zDetermine which FPN level each RoI in a set of RoIs should map to based on the heuristic in the FPN paper. Args: k_min (int) k_max (int) canonical_scale (int) canonical_level (int) eps (float) r%r&r'r(r)cJ||_||_||_||_||_yr+)r%r&s0lvl0r))selfr%r&r'r(r)s r"__init__zLevelMapper.__init__;s'  !# r$boxlistsrc Ptjtj|Dcgc] }t|c}}tj|j tj ||jz ztj|j|jz}tj||j|j}|jtj|jz jtjScc}w)z< Args: boxlists (list[BoxList]) r)minmax)rsqrtcatrfloorr1log2r0tensorr)rclampr%r&toint64)r2r4boxlists target_lvlss r"__call__zLevelMapper.__call__Is JJuyy8!L(7"3!LM Nkk$))ejjTWW.E"E UYU]U]efelelHm"mn kk+4::4::N u{{+djj8<) __name__ __module__ __qualname____doc__intfloatr3listrrDr$r"r,r,/sa  #          Jf J& Jr$r,boxesc ftj|d}|j|j}}tjt |Dcgc]6\}}tj |ddddf||tj |8c}}d}tj||gd}|Scc}}w)Nr)dimr )rlayoutr)rr:rr enumerate full_likestrided)rP concat_boxesrribidsroiss r"_convert_to_roi_formatr\Ws99U*L ''););EF ))dmnsdtu\`\]_`1bqb51E%--PV Wu  C 99c<(a 0D K vs ;B- feature original_sizec |jdd}g}t||D]j\}}t|t|z }dttj|j j z}|j|l|dS)Nrr)shapeziprMrr=r<roundappend)r]r^rpossible_scaless1s2 approx_scalescales r" _infer_scalerjbs == D#%OdM*&BRy59, U5<< 5::<BBDEEu%& 1 r$features image_shapesc$|s tdd}d}|D] }t|d|}t|d|}"||f}|Dcgc]}t||} }tjtj | dtj j } tjtj | dtj j } tt| t| ||} | | fScc}w)Nzimages list should not be emptyrr r6rr'r() ValueErrorr8rjrr<r=float32itemr-rL) rkrlr'r(max_xmax_yraoriginal_input_shapefeatscaleslvl_minlvl_max map_levelss r" _setup_scalesrzms :;; E E%E!He$E!He$%"5>CK L4l4!56 LF Lzz%,,vay FGLLNNGzz%,,vbzGHMMOOG  G  G '' J : MsD x featmap_namescfg}|jD]\}}||vs |j||Sr+)itemsrd)r{r| x_filteredkvs r" _filter_inputrs>J !1    a ! r$r output_sizesampling_ratiorvmapperc|| tdt|}t|}|dk(rt|d|||d|S||}t|} |djd} |dj |dj } } tj| | f|z| | } g}tt||D]\}\}}tj||k(d}||}t|||||}tjr!|j|j| n|j| j | |<tjr t!||} | S)a Args: x_filtered (List[Tensor]): List of input tensors. boxes (List[Tensor[N, 4]]): boxes to be used to perform the pooling operation, in (x1, y1, x2, y2) format and in the image reference size, not the feature map reference. The coordinate must satisfy ``0 <= x1 < x2`` and ``0 <= y1 < y2``. output_size (Union[List[Tuple[int, int]], List[int]]): size of the output sampling_ratio (int): sampling ratio for ROIAlign scales (Optional[List[float]]): If None, scales will be automatically inferred. Default value is None. mapper (Optional[LevelMapper]): If none, mapper will be automatically inferred. Default value is None. Returns: result (Tensor) z$scales and mapper should not be Noner r)r spatial_scalerr)rorr\r rarrrrrTrbr torchvision _is_tracingrdr?r#)rrPrrrvr num_levelsr[r num_rois num_channelsrrresulttracing_resultsr per_level_featureri idx_in_levelrois_per_levelresult_idx_in_levels r"_multiscale_roi_alignrs,~?@@ZJ !% (DQ qM # ))   E]F4yHa=&&q)LqM''A)=)=6E [[       FO-6s:v7N-OH))!5{{6U?3A6 l+'  #)    " " $  " "#6#9#9%#@ A$7#9#9&,,#GF< -H0 #FO< Mr$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]) )rvryrFrGrnr|rrr'r(ct|t|t|tr||f}||_||_t||_d|_ d|_ ||_ ||_ yr+) superr3r isinstancerLr|rtuplerrvryr'r()r2r|rrr'r( __class__s r"r3zMultiScaleRoIAlign.__init__ sg D! k3 '& 4K*, - ..r$r{rPrlrc,t||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) ) rr|rvryrzr'r(rrr)r2r{rPrlrs r"forwardzMultiScaleRoIAlign.forward!s&#1d&8&89 ;; $//"9+8L$*>*>@T@T, (DK%         KK OO   r$c|jjd|jd|jd|jdS)Nz(featmap_names=z, output_size=z, sampling_ratio=))rrHr|rr)r2s r"__repr__zMultiScaleRoIAlign.__repr__CsM~~&&'t7I7I6JK++,,=d>Q>Q=RRS U r$)rHrIrJrKrrNrMr,__annotations__strrrLrr3dictrrr __classcell__)rs@r"rrs!F"*$u+!6h{F[\O # /Cy/3c DI56/ /  //*  V   F|  5c?+    D # r$rrE)!typingrrrtorch.fxrrrtorchvision.ops.boxesrutilsr r jitunusedrNr#rLrMr-r,r\rjfxwraprrzrrrrModulerrOr$r"rs" *' vf&, L L LL L  L%J%JP$v,6&cu6l*.uS#X*?RUhk 4; #$6T#v+&tCyT&\PV P <PcP P T%[ ! P [ ! P PPfa a r$