:j2,VddlZddlmZddlZddlZddlmZmZddlmZddl m Z ddl m Z ddl mZmZdd lmZd d lmZmZd Zd ZdZeddZej2j4 ddedeeeefde ededededefdZGddej@Z!y)N)Union)nnTensor)is_compile_supported)BroadcastingList2)_pair)_assert_has_ops_has_ops)_log_api_usage_once)check_roi_boxes_shapeconvert_boxes_to_roi_formatc fd}|S)zkLazily wrap a function with torch.compile on the first call This avoids eagerly importing dynamo. cFtjfd}|S)Nctjfi}tj|t j <||i|SN)torchcompile functoolswrapsglobals__name__)argskwargs compiled_fncompile_kwargsfns n/home/mcse/projects/srt_converter/srt-converter-venv/lib/python3.12/site-packages/torchvision/ops/roi_align.py compile_hookz7lazy_compile..decorate_fn..compile_hooksF--=n=K%8Y__R%8%EGIbkk "// /)rr)rr rs` r decorate_fnz!lazy_compile..decorate_fns&   0  0 r!)rr"s` r lazy_compiler$s  r!cj\}}}|jd}|jd}|j} |j} tj| |dz k\|dz | dz} tj| |dz k\|dz | } tj| |dz k\|j j |}tj| |dz k\|dz | dz} tj| |dz k\|dz | } tj| |dz k\|j j |}|| z } || z }d| z }d|z }fd}|| | }|| | }|| | }|| | }d}|||}|||}|| |}|| |}||z||zz||zz||zz}|S)Nrminr ?c LFJtjdddddf|d}tjdddddf|d}dddddddftjjdddddddf|dddddddddf|dddddddddffS)Nrdevice)rwherearanger+)yxchannelsinput roi_batch_indxmaskymasks r masked_indexz+_bilinear_interpolate..masked_indexBs  $ $$ E!T1*-q!4A E!T1*-q!4A !T4tT9 : LL%,, 7atTSW8W X aq$4' ( atQa' ( *  r!c H|dddddddddf|dddddddddfzSrr#)r.r/s r outer_prodz)_bilinear_interpolate..outer_prodWs1D!T1d*+a4q$0I.JJJr!)sizeclampintrr,todtype)r1r2r.r/r4r3_heightwidthy_lowx_lowy_highx_highlylxhyhxr5v1v2v3v4r7w1w2w3w4valr0s`` `` @r_bilinear_interpolaterQ#s"'Ax AA AA EEGE EEGE [[&1*,fqj%!) DF KK!+VaZ ?E EVaZ'ekk):A>A [[%!)+UQY BF KK*EAIu =E EUQY&U[[(91=A UB UB rB rB    eU #B eV $B fe $B ff %BK B B B B B B B B r'BG b2g %R /C Jr!ctjr9|jr-|jtjk7r|j S|Sr)ris_autocast_enabledis_cudar<doublefloat)tensors r maybe_castrXes4   "v~~&,,%,,:V||~ r!T)dynamicc |j}t|}t|}|j\}}} } tj||j } tj||j } |dddfj } |rdnd}|dddf|z|z }|dddf|z|z }|dddf|z|z }|dddf|z|z }||z }||z }|s.tj|d }tj|d }||z }||z }|dkD}|r|ntj||z }|r|ntj||z } |rVt||zd}tj||j }tj||j }d}d}ntj||zd }tj| |j }tj| |j }|dddf|dddfk}|dddf|dddfk}d }||| ddddf||zz|ddddfdzj|j|||z zz} ||| ddddf||zz|ddddfdzj|j|||z zz}!t|| | |!||}"|sHtj|ddddddddf|"d}"tj|ddddddddf|"d}"|"jd }#t|tjr|#|dddddfz}#n|#|z}#|#j|}#|#S) Nr*rg?gr r r(r&c|ddddfSrr#)ts rfrom_Kz_roi_align..from_KsD$r!))r<rXr8rr-r+r:r9ceilmaxr;rQr,sum isinstancer)$r1rois spatial_scale pooled_height pooled_widthsampling_ratioaligned orig_dtyper=r>r?phpwr2offset roi_start_w roi_start_h roi_end_w roi_end_h roi_width roi_height bin_size_h bin_size_wexact_samplingroi_bin_grid_hroi_bin_grid_wcountiyixr4r3r_r.r/rPoutputs$ r _roi_alignrrsJ u E d D**,Aq&% mELL 9B l5<< 8B AJNN$MSFq!t*},v5Kq!t*},v5KQT ]*V3IQT ]*V3IK'I[(J KK s3 [[5 m+J\)J#a'N'5^5::jS`F`;aN'5^5::iR^F^;_NN^3Q7 \\. > \\. > N^;C \\& 6 \\% 547 nQW5547 nQW55  { T1d] fZ0 0 1 dD!m s " & &u{{ 3fZ.=X6Y Y Z { T1d] fZ0 0 1 dD!m s " & &u{{ 3fZ.=X6Y Y Z }aE5 IC kk%4tQ <=sAFkk%4tT1 <=sAF WWX F%&%4t+,,% YYz "F Mr!r1boxes output_sizergrjrkreturnc tjjs-tjjst t t ||}t|}t|tjs t|}tjjsxtr8tjrZ|js|js |jr6t!|j"j$rt'||||d|d||St)tj*j,j ||||d|d||S)aj Performs Region of Interest (RoI) Align operator with average pooling, as described in Mask R-CNN. Args: input (Tensor[N, C, H, W]): The input tensor, i.e. a batch with ``N`` elements. Each element contains ``C`` feature maps of dimensions ``H x W``. If the tensor is quantized, we expect a batch size of ``N == 1``. boxes (Tensor[K, 5] or List[Tensor[L, 4]]): the box coordinates in (x1, y1, x2, y2) format where the regions will be taken from. The coordinate must satisfy ``0 <= x1 < x2`` and ``0 <= y1 < y2``. If a single Tensor is passed, then the first column should contain the index of the corresponding element in the batch, i.e. a number in ``[0, N - 1]``. If a list of Tensors is passed, then each Tensor will correspond to the boxes for an element i in the batch. output_size (int or Tuple[int, int]): the size of the output (in bins or pixels) after the pooling is performed, as (height, width). spatial_scale (float): a scaling factor that maps the box coordinates to the input coordinates. For example, if your boxes are defined on the scale of a 224x224 image and your input is a 112x112 feature map (resulting from a 0.5x scaling of the original image), you'll want to set this to 0.5. Default: 1.0 sampling_ratio (int): number of sampling points in the interpolation grid used to compute the output value of each pooled output bin. If > 0, then exactly ``sampling_ratio x sampling_ratio`` sampling points per bin are used. If <= 0, then an adaptive number of grid points are used (computed as ``ceil(roi_width / output_width)``, and likewise for height). Default: -1 aligned (bool): If False, use the legacy implementation. If True, pixel shift the box coordinates it by -0.5 for a better alignment with the two neighboring pixel indices. This version is used in Detectron2 Returns: Tensor[K, C, output_size[0], output_size[1]]: The pooled RoIs. rr )rjit is_scripting is_tracingr roi_alignrrrerrr $are_deterministic_algorithms_enabledrTis_mpsis_xpurr+typerr ops torchvision)r1rrrgrjrkrfs rrrsR 99 ! ! #EII,@,@,BI&% D $K dELL )*40 99 ! ! # ::<%--SXS_S_chcoco"5<<#4#45eT=+a.+VW.Zhjqr r 99 * * t]KNKNNT[ r!c neZdZdZ d deedededeffd Zde de e e e fd e fd Z d e fd ZxZS) RoIAlignz See :func:`roi_align`. rrgrjrkcpt|t|||_||_||_||_yr)super__init__r rrgrjrk)selfrrgrjrk __class__s rrzRoIAlign.__init__ s7 D!&*, r!r1rfrcrt|||j|j|j|jSr)rrrgrjrk)rr1rfs rforwardzRoIAlign.forwards0d&6&68J8JDL_L_aeamamnnr!c |jjd|jd|jd|jd|j d }|S)Nz (output_size=z, spatial_scale=z, sampling_ratio=z , aligned=))rrrrgrjrk)rss r__repr__zRoIAlign.__repr__s^~~&&'(++,t112 3 34   r!)F)r __module__ __qualname____doc__rr:rVboolrrrlistrstrr __classcell__)rs@rrrsu &s+     oVo5f1E+Fo6o # r!r)r(r`F)"rtypingrrtorch.fxrrtorch._dynamo.utilsrtorch.jit.annotationsrtorch.nn.modules.utilsrtorchvision.extensionr r utilsr _utilsrrr$rQrXrfxwraprr:rVrrModulerr#r!rrs 43(;'F&=DdUUp  8 8 f% &8#3'8 8  8  8 88vryyr!