o `)i8,@sddlZddlmZmZddlZddlZddlmZmZddlm Z ddl m Z ddl m Z ddlmZmZdd lmZd d lmZmZd d ZddZddZeddddZejj   d$dedeeeefde ededededefd d!ZGd"d#d#ej Z!dS)%N)ListUnion)nnTensor)is_compile_supported)BroadcastingList2)_pair)_assert_has_ops_has_ops)_log_api_usage_once)check_roi_boxes_shapeconvert_boxes_to_roi_formatc sfdd}|S)zkLazily wrap a function with torch.compile on the first call This avoids eagerly importing dynamo. cstfdd}|S)Ncs6tjfi}t|tj<||i|SN)torchcompile functoolswrapsglobals__name__)argskwargsZ compiled_fn)compile_kwargsfne/home/app/PaddleOCR-VL-test/.venv_paddleocr/lib/python3.10/site-packages/torchvision/ops/roi_align.py compile_hooksz7lazy_compile..decorate_fn..compile_hook)rr)rrr)rr decorate_fnsz!lazy_compile..decorate_fnr)rrrrr lazy_compiles  r cs\}}}|jdd}|jdd}|} |} t| |dk|d| d} t| |dk|d| } t| |dk|j|}t| |dk|d| d} t| |dk|d| } t| |dk|j|}|| } || }d| }d|}fdd}|| | }|| | }|| | }|| | }dd}|||}|||}|| |}|| |}||||||||}|S) Nrminr ?c sdur*dus Jtdddddf|d}tdddddf|d}dddddddftjjddddddddf|dddddddddf|dddddddddffS)Nrdevice)rwherearanger%yxZchannelsinput roi_batch_indxmaskymaskrr masked_indexBs   "z+_bilinear_interpolate..masked_indexc Ss@|dddddddddf|dddddddddfSrrr(rrr outer_prodWs@z)_bilinear_interpolate..outer_prod)sizeclampintrr&todtype)r,r-r)r*r/r._heightwidthZy_lowZx_lowZy_highZx_highZlylxhyhxr0v1v2Zv3Zv4r1Zw1Zw2Zw3Zw4valrr+r_bilinear_interpolate#s6           r@cCs&tr|jr|jtjkr|S|Sr)rZis_autocast_enabledis_cudar6doublefloat)Ztensorrrr maybe_castesrDT)Zdynamicc$ Cs6|j}t|}t|}|\}}} } tj||jd} tj||jd} |dddf} |r1dnd}|dddf||}|dddf||}|dddf||}|dddf||}||}||}|s{tj|d d }tj|d d }||}||}|dk}|r|nt||}|r|nt||} |rt ||d}tj||jd}tj||jd}d}d}n9tj||dd }tj| |jd}tj| |jd}|dddf|dddfk}|dddf|dddfk}d d }||| ddddf|||ddddfd |j|||} ||| ddddf|||ddddfd |j|||}!t || | |!||}"|swt |ddddddddf|"d}"t |ddddddddf|"d}"|" d }#t|tjr|#|dddddf}#n|#|}#|# |}#|#S)Nr$rg?gr r r#r!cSs|ddddfSrr)trrrfrom_Ksz_roi_align..from_K))r6rDr2rr'r%r4r3ceilmaxr5r@r&sum isinstancer)$r,rois spatial_scaleZ pooled_heightZ pooled_widthsampling_ratioalignedZ orig_dtyper7r8r9phpwr-offsetZ roi_start_wZ roi_start_hZ roi_end_wZ roi_end_hZ roi_widthZ roi_heightZ bin_size_hZ bin_size_wZexact_samplingZroi_bin_grid_hZroi_bin_grid_wcountZiyZixr/r.rHr)r*r?outputrrr _roi_alignrsr   ((&&  rXr#rIFr,boxes output_sizerPrQrRreturnc Cstjstjsttt||}t|}t|tj s"t |}tjsKt r7t rK|j s7|js7|jrKt|jjrKt||||d|d||Sttjj||||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 )rZjitZ is_scripting is_tracingr roi_alignrrrNrrr Z$are_deterministic_algorithms_enabledrAZis_mpsZis_xpurr%typerXr opsZ torchvision)r,rYrZrPrQrRrOrrrr]s2)   r]c sjeZdZdZ ddeedededeffdd Zd e d e e e e fd e fd d Z d e fddZZS)RoIAlignz See :func:`roi_align`. FrZrPrQrRcs.tt|||_||_||_||_dSr)super__init__r rZrPrQrR)selfrZrPrQrR __class__rrrb s  zRoIAlign.__init__r,rOr[cCst|||j|j|j|jSr)r]rZrPrQrR)rcr,rOrrrforwardszRoIAlign.forwardc Cs2|jjd|jd|jd|jd|jd }|S)Nz (output_size=z, spatial_scale=z, sampling_ratio=z , aligned=))rerrZrPrQrR)rcsrrr__repr__s zRoIAlign.__repr__)F)r __module__ __qualname____doc__rr4rCboolrbrrrrfstrri __classcell__rrrdrr`s "r`)r#rIF)"rtypingrrrZtorch.fxrrZtorch._dynamo.utilsrZtorch.jit.annotationsrZtorch.nn.modules.utilsrZtorchvision.extensionr r utilsr _utilsrrr r@rDrXZfxwrapr4rCrmr]Moduler`rrrrsH    B X ;