a d3 @s@ddlZddlmZddlmZddlmZmZmZm Z m Z m Z ddl Z ddl m Z mZddlmZddlmZdd lmZd d lmZd d lmZmZmZd d lmZd dlmZmZd dl m!Z!m"Z"ddlmZ#ddl$m%Z%ddl&m'Z'ddl(m)Z)m*Z*ddgZ+e,e,e,ede j-fe j.dddZ/e,e,ede j-fe j.dddZ0e j-ddd Z1Gd!d"d"e j-Z2Gd#d$d$e*Z3Gd%d&d&e*Z4Gd'd(d(e j-Z5e ej6ej7fe,ede j-fd)d*d+Z8Gd,ddeZ9eed-e9j:fd.e"j;fd/dd0de"j;ddd1e e9eRs zSSDLiteHead.__init__xr&cCs||||dS)N)Zbbox_regression cls_logits)rBr@)rCrGr,r,r-forwardYszSSDLiteHead.forward)__name__ __module__ __qualname__rintrr Moduler>r rstrrI __classcell__r,r,rDr-r8Qsr8cs:eZdZeeeeeedejfdfdd ZZ S)r?.r9csNt}t||D] \}}|t|||d|qt|t||dS)Nr r ModuleListzipappendr.r7r=r>)rCr"r:r;r%rHchannelsanchorsrDr,r-r>as z"SSDLiteClassificationHead.__init__ rJrKrLrrMrr rNr>rPr,r,rDr-r?`sr?cs8eZdZeeeeedejfdfdd ZZ S)rA.)r"r:r%csNt}t||D] \}}|t|d|d|qt|t|ddS)Nr rQ)rCr"r:r%Zbbox_regrUrVrDr,r-r>ls zSSDLiteRegressionHead.__init__rWr,r,rDr-rAksrAcsPeZdZd ejeedejfeedfdd Ze e e e fddd Z Z S) SSDLiteFeatureExtractorMobileNet?.)backbonec4_posr% width_mult min_depthc stt|||jr$tdttjg|d|||jdRtj||jddg||ddR|_fdd}t t |dj |d|t |d|d|t |d|d|t |d|d |g}t |||_ dS) Nz0backbone[c4_pos].use_res_connect should be Falserrcstt|Sr<)maxrM)dr_r^r,r-z;SSDLiteFeatureExtractorMobileNet.__init__..i)r=r>rZuse_res_connect ValueErrorr r)blockfeaturesrRr/r#r7extra)rCr\r]r%r^r_Z get_depthrkrDrbr-r>us$  $*z)SSDLiteFeatureExtractorMobileNet.__init__rFcCsVg}|jD]}||}||q |jD]}||}||q(tddt|DS)NcSsg|]\}}t||fqSr,)rO).0ivr,r,r- rdz.)rjrTrkr enumerate)rCrGoutputrir,r,r-rIs    z(SSDLiteFeatureExtractorMobileNet.forward)rZr[)rJrKrLr rNrMrfloatr>r rrOrIrPr,r,rDr-rYts !rY)r\trainable_layersr%cCs|j}dgddt|Dt|dg}t|}d|krH|ksRntd|dkrbt|n |||}|d|D]}|D]}|dqqzt||d|S)NrcSs g|]\}}t|ddr|qS)Z_is_cnF)getattr)rlrmbr,r,r-rordz(_mobilenet_extractor..rzYtrainable_layers should be in the range [0, {num_stages}], instead got {trainable_layers}F)rjrplenrh parametersZrequires_grad_rY)r\rsr%Z stage_indicesZ num_stagesZ freeze_beforeruZ parameterr,r,r-_mobilenet_extractors& ryc @s8eZdZedededddddiidd d d d ZeZd S)r zShttps://download.pytorch.org/models/ssdlite320_mobilenet_v3_large_coco-a79551df.pthi}4)rrz]https://github.com/pytorch/vision/tree/main/references/detection#ssdlite320-mobilenetv3-largez COCO-val2017Zbox_mapgL5@g-?gt*@zSThese weights were produced by following a similar training recipe as on the paper.)Z num_params categoriesZmin_sizeZrecipeZ_metricsZ_ops _file_sizeZ_docs)urlZ transformsmetaN)rJrKrLrrrCOCO_V1DEFAULTr,r,r,r-r s"Z pretrainedZpretrained_backbone)weightsweights_backboneT)rprogressr;rtrainable_backbone_layersr%)rrr;rrr%kwargsr&cKst|}t|}d|vr&td|durJd}td|t|jd}n |durVd}t|dupf|du|dd}|du}|durt t j dd d }t f||||d |}|durt |t|||}d } td dtdDddd} t|| } | } t| t| jkr*tdt| dt| jddddgdgdd} i| |}t|| | |fdt| | ||i|}|dur||j|d|S)aSSDlite model architecture with input size 320x320 and a MobileNetV3 Large backbone, as described at `Searching for MobileNetV3 `__ and `MobileNetV2: Inverted Residuals and Linear Bottlenecks `__. .. betastatus:: detection module See :func:`~torchvision.models.detection.ssd300_vgg16` for more details. Example: >>> model = torchvision.models.detection.ssdlite320_mobilenet_v3_large(weights=SSDLite320_MobileNet_V3_Large_Weights.DEFAULT) >>> model.eval() >>> x = [torch.rand(3, 320, 320), torch.rand(3, 500, 400)] >>> predictions = model(x) Args: weights (:class:`~torchvision.models.detection.SSDLite320_MobileNet_V3_Large_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.detection.SSDLite320_MobileNet_V3_Large_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. num_classes (int, optional): number of output classes of the model (including the background). weights_backbone (:class:`~torchvision.models.MobileNet_V3_Large_Weights`, optional): The pretrained weights for the backbone. trainable_backbone_layers (int, optional): number of trainable (not frozen) layers starting from final block. Valid values are between 0 and 6, with 6 meaning all backbone layers are trainable. If ``None`` is passed (the default) this value is set to 6. norm_layer (callable, optional): Module specifying the normalization layer to use. **kwargs: parameters passed to the ``torchvision.models.detection.ssd.SSD`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.detection.SSDLite320_MobileNet_V3_Large_Weights :members: sizez?The size of the model is already fixed; ignoring the parameter.Nr;rz[gMbP?r1)ZepsZmomentum)rrr%Z reduced_tail)@rcSsg|] }ddgqS)rr r,)rl_r,r,r-ro,rdz1ssdlite320_mobilenet_v3_large..g?gffffff?)Z min_ratioZ max_ratioz4The length of the output channels from the backbone z? do not match the length of the anchor generator aspect ratios g?i,)?rr)Z score_threshZ nms_threshZdetections_per_imgZtopk_candidatesZ image_meanZ image_stdhead)r)r verifyrwarningswarnrrwr}rrr Z BatchNorm2drr7ryrrange det_utilsZretrieve_out_channelsZnum_anchors_per_locationZ aspect_ratiosrhrr8Zload_state_dictZget_state_dict)rrr;rrr%rZ reduce_tailr\rZanchor_generatorr#r:defaultsmodelr,r,r-r!sp8       )>r collectionsr functoolsrtypingrrrrrr r5r r Zops.miscr Ztransforms._presetsrutilsrrZ_apirrrZ_metarrrrZ mobilenetv3rrrZ anchor_utilsrZbackbone_utilsrZssdrr__all__rMrNr)r.r/r7r8r?rArYZ MobileNetV2Z MobileNetV3ryr r~Z IMAGENET1K_V1boolr!r,r,r,r-sn               1