o V h.@sddlmZmZmZmZmZddlZddlZddlZddlm Z m Z ddl m Z ddl mZddlmZejjd e d ee d e fd d Z   d2dededededef ddZGdddZdee d e fddZde deed efdd Zejjd!ee d"eeeefdeded eeeeff d#d$Zejjd%eee fd&eed ee fd'd(Zejjd)ee dee d*eed+ed,eeed-eed e fd.d/ZGd0d1d1e jZ dS)3)DictListOptionalTupleUnionN)nnTensorbox_area)_log_api_usage_once) roi_alignlevelsunmerged_resultsreturnc Cs|d}|j|j}}tj|d|d|d|df||d}tt|D]4}t||kddddd}| |d||d||d||d}| d|||}q)|S)Nrr r dtypedevice) rrtorchzerossizerangelenwhereviewexpandscatter)rr first_resultrrreslevelindexr$K/var/www/vscode/kcb/lib/python3.10/site-packages/torchvision/ops/poolers.py_onnx_merge_levelss&   r&ư>k_mink_maxcanonical_scalecanonical_levelepscCst|||||SN) LevelMapper)r*r+r,r-r.r$r$r%initLevelMapper%sr1c @sLeZdZdZ   ddedededed ef d d Zd eed efddZ dS)r0zDetermine 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-r.cCs"||_||_||_||_||_dSr/)r*r+s0lvl0r.)selfr*r+r,r-r.r$r$r%__init__;s  zLevelMapper.__init__boxlistsrcCsvttdd|D}t|jt||jtj|j|j d}tj ||j |j d}| tj|j  tjS)z< Args: boxlists (list[BoxList]) cSsg|]}t|qSr$r ).0boxlistr$r$r% Osz(LevelMapper.__call__..r)minmax)rsqrtcatfloorr3log2r2tensorr.rclampr*r+toint64)r4r6s target_lvlsr$r$r%__call__Is.zLevelMapper.__call__Nr'r(r)) __name__ __module__ __qualname____doc__intfloatr5rrrGr$r$r$r%r0/s" r0boxescsTtj|dd}|j|jtjfddt|Ddd}tj||gdd}|S)Nr)dimc s6g|]\}}tj|ddddf|tjdqS)Nr )rlayoutr)r full_likestrided)r7ibrrr$r%r9[s6z*_convert_to_roi_format..r )rr>rr enumerate)rO concat_boxesidsroisr$rVr%_convert_to_roi_formatWsr[feature original_sizecCsb|jdd}g}t||D]\}}t|t|}dtt|}||q|dS)Nr r)shapeziprNrrAr@roundappend)r\r]rpossible_scaless1s2 approx_scalescaler$r$r% _infer_scalebs rhfeatures image_shapesc s|stdd}d}|D]}t|d|}t|d|}q ||ffdd|D}ttj|dtjd }ttj|dtjd } tt|t| ||d} || fS) Nzimages list should not be emptyrr csg|]}t|qSr$)rh)r7featoriginal_input_shaper$r%r9zsz!_setup_scales..r:rr,r-) ValueErrorr<rr@rAfloat32itemr1rM) rirjr,r-max_xmax_yr_scaleslvl_minlvl_max map_levelsr$rlr% _setup_scalesms$  rxx featmap_namescCs,g}|D] \}}||vr||q|Sr/)itemsrb)ryrz x_filteredkvr$r$r% _filter_inputs  rr| output_sizesampling_ratiortmappercCs"|dus|dur tdt|}t|}|dkr%t|d|||d|dS||}t|} |djd} |dj|dj} } tj| | f|| | d} g}t t ||D]1\}\}}t ||kd}||}t|||||d}t r}||| qT|| j| |<qTt 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$scales and mapper should not be Noner r)r spatial_scalerr)rorr[rr_rrrrrWr`r torchvision _is_tracingrbrCr&)r|rOrrrtr num_levelsrZrnum_rois num_channelsrrresulttracing_resultsr"per_level_featurerg idx_in_levelrois_per_levelresult_idx_in_levelr$r$r%_multiscale_roi_alignsT   rc seZdZdZeeeeedZddddee de e e e ee fde d e d e f fd d Z d ee efdeedee e e fdefddZde fddZZS)MultiScaleRoIAligna{ Multi-scale RoIAlign pooling, which is useful for detection with or without FPN. It infers the scale of the pooling via the heuristics specified in eq. 1 of the `Feature Pyramid Network paper `_. They keyword-only parameters ``canonical_scale`` and ``canonical_level`` correspond respectively to ``224`` and ``k0=4`` in eq. 1, and have the following meaning: ``canonical_level`` is the target level of the pyramid from which to pool a region of interest with ``w x h = canonical_scale x canonical_scale``. Args: featmap_names (List[str]): the names of the feature maps that will be used for the pooling. output_size (List[Tuple[int, int]] or List[int]): output size for the pooled region sampling_ratio (int): sampling ratio for ROIAlign canonical_scale (int, optional): canonical_scale for LevelMapper canonical_level (int, optional): canonical_level for LevelMapper Examples:: >>> m = torchvision.ops.MultiScaleRoIAlign(['feat1', 'feat3'], 3, 2) >>> i = OrderedDict() >>> i['feat1'] = torch.rand(1, 5, 64, 64) >>> i['feat2'] = torch.rand(1, 5, 32, 32) # this feature won't be used in the pooling >>> i['feat3'] = torch.rand(1, 5, 16, 16) >>> # create some random bounding boxes >>> boxes = torch.rand(6, 4) * 256; boxes[:, 2:] += boxes[:, :2] >>> # original image size, before computing the feature maps >>> image_sizes = [(512, 512)] >>> output = m(i, [boxes], image_sizes) >>> print(output.shape) >>> torch.Size([6, 5, 3, 3]) )rtrwr'r(rnrzrrr,r-csVtt|t|tr||f}||_||_t||_d|_ d|_ ||_ ||_ dSr/) superr5r isinstancerMrzrtuplerrtrwr,r-)r4rzrrr,r- __class__r$r%r5 s    zMultiScaleRoIAlign.__init__ryrOrjrcCsTt||j}|jdus|jdurt|||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) N) rrzrtrwrxr,r-rrr)r4ryrOrjr|r$r$r%forward!s   zMultiScaleRoIAlign.forwardcCs&|jjd|jd|jd|jdS)Nz(featmap_names=z, output_size=z, sampling_ratio=))rrIrzrr)r4r$r$r%__repr__Cs zMultiScaleRoIAlign.__repr__)rIrJrKrLrrrNr0__annotations__strrrMrr5rrrr __classcell__r$r$rr%rs4#  "rrH)!typingrrrrrrtorch.fxrrrtorchvision.ops.boxesr utilsr rjitunusedr&rMrNr1r0r[rhfxwraprxrrrModulerr$r$r$r%st     (   (  S