^jN dZddlZddlmZddlmZmZmZddlZddl m Z ddl m cm Z ddlmZmZddlmZddlmZdd lmZmZd gZd ZGd d e j4ZGdde j4ZGddeZGddeZGddeZGdde j4Z Gdd e j4Z!d2dZ"d3dZ#ee#de#dde#d e#d!e#d"e#d#e#d$de#d%e#d&d' Z$ed2d(e!fd)Z%ed2d(e!fd*Z&ed2d(e!fd+Z'ed2d(e!fd,Z(ed2d(e!fd-Z)ed2d(e!fd.Z*ed2d(e!fd/Z+ed2d(e!fd0Z,ed2d(e!fd1Z-y)4a: SEResNet implementation from Cadene's pretrained models https://github.com/Cadene/pretrained-models.pytorch/blob/master/pretrainedmodels/models/senet.py Additional credit to https://github.com/creafz Original model: https://github.com/hujie-frank/SENet ResNet code gently borrowed from https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py FIXME I'm deprecating this model and moving them to ResNet as I don't want to maintain duplicate support for extras like dilation, switchable BN/activations, feature extraction, etc that don't exist here. N) OrderedDict)TypeOptionalTupleIMAGENET_DEFAULT_MEANIMAGENET_DEFAULT_STD)create_classifier)build_model_with_cfg)register_modelgenerate_default_cfgsSENetcpt|tjr-tjj |j ddyt|tj rUtjj|j dtjj|jdyy)Nfan_outrelu)mode nonlinearityg?) isinstancennConv2dinitkaiming_normal_weight BatchNorm2d constant_bias)ms \/var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/timm/models/senet.py _weight_initr!sp!RYY yvN Ar~~ & !((B' !&&"% 'c0eZdZddedeffd ZdZxZS)SEModulechannels reductionc||d}t|tj|||zfddi||_tj d|_tj||z|fddi||_tj|_ y)Ndevicedtype kernel_sizer Tinplace) super__init__rrfc1ReLUrfc2Sigmoidsigmoid)selfr%r&r)r*dd __class__s r r/zSEModule.__init__(sw / 99Xx9'<R!RrRGGD) 99X2HR!RrRzz| r"c|}|jdd}|j|}|j|}|j|}|j |}||zS)N)T)keepdim)meanr0rr2r4)r5x module_inputs r forwardzSEModule.forward0sX FF64F ( HHQK IIaL HHQK LLOar")NN)__name__ __module__ __qualname__intr/r? __classcell__r7s@r r$r$&s$$$ r"r$ceZdZdZdZy) BottleneckzH Base class for bottlenecks that implements `forward()` method. c|}|j|}|j|}|j|}|j|}|j |}|j|}|j |}|j |}|j|j|}|j||z}|j|}|SN) conv1bn1rconv2bn2conv3bn3 downsample se_moduler5r=shortcutouts r r?zBottleneck.forward?sjjmhhsmiinjjohhsmiinjjohhsm ?? &q)HnnS!H,iin r"N)r@rArB__doc__r?r"r rGrG:s r"rGcdeZdZdZdZ d dedededededeejf fd Z xZ S) SEBottleneckz" Bottleneck for SENet154. inplanesplanesgroupsr&striderPc <||d} t |tj||dzfddd| |_tj |dzfi| |_tj|dz|dzfd|d|dd| |_tj |dzfi| |_tj|dz|dzfddd| |_ tj |dzfi| |_ tjd |_ t|dzfd |i| |_||_||_y) Nr(r9r Fr+rrYr:r+r]paddingr\rTr,r&r.r/rrrJrrKrLrMrNrOr1rr$rQrPr] r5rZr[r\r&r]rPr)r*r6r7s r r/zSEBottleneck.__init__\s' / YYx!UURTU >>&1*33YY QJ QJ       >>&1*33YYvz6A:W15WTVW >>&1*33GGD) !&1*H HRH$ r"r NNN r@rArBrU expansionrCrrModuler/rDrEs@r rXrXVsjI.2     !+r"rXcdeZdZdZdZ d dedededededeejf fd Z xZ S) SEResNetBottleneckz ResNet bottleneck with a Squeeze-and-Excitation module. It follows Caffe implementation and uses `stride=stride` in `conv1` and not in `conv2` (the latter is used in the torchvision implementation of ResNet). rYrZr[r\r&r]rPc ||d} t |tj||fdd|d| |_tj |fi| |_tj||fdd|dd| |_tj |fi| |_tj||dzfddd| |_ tj |dzfi| |_ tjd |_ t|dzfd |i| |_||_||_y) Nr(r Fr+rr]r:r+rar\rrYr_Tr,r&rbrcs r r/zSEResNetBottleneck.__init__s / YYx`QUSY`]_` >>&/B/YYvvi1aPV]bifhi >>&/B/YYvvzSquSPRS >>&1*33GGD) !&1*H HRH$ r"rdrerEs@r riri~sj I.2     !+r"ricjeZdZdZdZ d dedededededeejd effd Z xZ S) SEResNeXtBottleneckzI ResNeXt bottleneck type C with a Squeeze-and-Excitation module. rYrZr[r\r&r]rP base_widthc V|| d} t |tj||dz z|z} t j || fdddd| |_t j| fi| |_t j | | fd|d|dd| |_ t j| fi| |_ t j | |dzfddd | |_ t j|dzfi| |_ t jd |_t|dzfd |i| |_||_||_y) Nr(@r Frkr:r`rYr_Tr,r&)r.r/mathfloorrrrJrrKrLrMrNrOr1rr$rQrPr]) r5rZr[r\r&r]rPror)r*r6widthr7s r r/zSEResNeXtBottleneck.__init__s /  6Z"_56?YYxZAERSZWYZ >>%.2.YYuev6ST]cjovsuv >>%.2.YYufqjRaeRrR >>&1*33GGD) !&1*H HRH$ r")r NrYNNrerEs@r rnrnswI.2     !+r"rncfeZdZdZ d dedededededeejf fd Zd Z xZ S) SEResNetBlockr rZr[r\r&r]rPc ||d} t |tj||fdd|dd| |_tj |fi| |_tj||fdd|dd| |_tj |fi| |_tjd|_ t|fd |i| |_ ||_ ||_y) Nr(r:r F)r+rar]rrlTr,r&)r.r/rrrJrrKrLrMr1rr$rQrPr]rcs r r/zSEResNetBlock.__init__s / YYxkQRX_dkhjk >>&/B/YYvvi1aPV]bifhi >>&/B/GGD) !&DIDD$ r"cZ|}|j|}|j|}|j|}|j|}|j |}|j|}|j |j |}|j ||z}|j|}|SrI)rJrKrrLrMrPrQrRs r r?zSEResNetBlock.forwardsjjmhhsmiinjjohhsmiin ?? &q)HnnS!H,iin r"rd) r@rArBrfrCrrrgr/r?rDrEs@r rvrvscI.2     !+,r"rvceZdZ ddeej deedfdededededed e d ed ed ed e ffd Z ddZ e jjddZe jjddZe jjdej fdZdd ed e fdZdZdde fdZdZxZS)rblocklayers.r\r& drop_ratein_chansrZ input_3x3downsample_kernel_sizedownsample_padding num_classes global_poolct|| |d}||_| |_||_||_|rdt j|ddfdddd|fd t jd)i|fd t jd fd t jddddd|fdt jd)i|fdt jd fdt jd|dfdddd|fdt j|fi|fdt jd fg }nMdt j||fddddd|fd t j|fi|fd t jd fg}t jt||_ t jddd |_t|ddg|_|j"|fd|d||ddd||_|xj td|j&zddgz c_|j"|fd|dd||| | d||_|xj td|j&zdd gz c_|j"|fd!|dd||| | d||_|xj td!|j&zd"d#gz c_|j"|fd$|dd||| | d||_|xj td$|j&zd%d&gz c_d$|j&zx|_|_t3|j.|jfd'| i|\|_|_|j9D] }t;|y()*af Parameters ---------- block (nn.Module): Bottleneck class. - For SENet154: SEBottleneck - For SE-ResNet models: SEResNetBottleneck - For SE-ResNeXt models: SEResNeXtBottleneck layers (list of ints): Number of residual blocks for 4 layers of the network (layer1...layer4). groups (int): Number of groups for the 3x3 convolution in each bottleneck block. - For SENet154: 64 - For SE-ResNet models: 1 - For SE-ResNeXt models: 32 reduction (int): Reduction ratio for Squeeze-and-Excitation modules. - For all models: 16 dropout_p (float or None): Drop probability for the Dropout layer. If `None` the Dropout layer is not used. - For SENet154: 0.2 - For SE-ResNet models: None - For SE-ResNeXt models: None inplanes (int): Number of input channels for layer1. - For SENet154: 128 - For SE-ResNet models: 64 - For SE-ResNeXt models: 64 input_3x3 (bool): If `True`, use three 3x3 convolutions instead of a single 7x7 convolution in layer0. - For SENet154: True - For SE-ResNet models: False - For SE-ResNeXt models: False downsample_kernel_size (int): Kernel size for downsampling convolutions in layer2, layer3 and layer4. - For SENet154: 3 - For SE-ResNet models: 1 - For SE-ResNeXt models: 1 downsample_padding (int): Padding for downsampling convolutions in layer2, layer3 and layer4. - For SENet154: 1 - For SE-ResNet models: 0 - For SE-ResNeXt models: 0 num_classes (int): Number of outputs in `last_linear` layer. - For all models: 1000 r(rJrqr:r9r F)r]rarrKrelu1Tr,rL)rqrqr:rMrelu2rNrOrelu3r+r]rar)r] ceil_modelayer0)num_chsr&moduler)r[blocksr\r&rrrYlayer1)r[rr]r\r&rrlayer2layer3i layer4 pool_typeN)rq)r.r/rZrr}r|rrrr1 Sequentialrr MaxPool2dpool0dict feature_info _make_layerrrfrrr num_featureshead_hidden_sizer r last_linearmodulesr!)r5rzr{r\r&r|r}rZr~rrrrr)r*r6layer0_modulesrr7s r r/zSENet.__init__sx  /  &  " "))Hb![Aqu[XZ[\0R01"''$/0"))UaURTUV0R01"''$/0"))B![Aqu[XZ[\x6267"''$/0 N"))HhmAaYZafmjlmnx6267"''$/0N mmK$?@ \\!A> !(aQR&d&&   !9#$       d2+?1U]^__&d&&   !9#91      d3+@AV^_``&d&&   !9#91      d3+@BW_`aa&d&&   !9#91      d3+@BW_`aa47%//4IID1->      . ".  . *$* A O r"c  | | d} d} |dk7s|j||jzk7rhtjtj|j||jzf|||dd| tj ||jzfi| } ||j||||| fi| g} ||jz|_t d|D]'}| j||j|||fi| )tj| S)Nr(r Fr)rZrfrrrrrangeappend)r5rzr[rr\r&r]rrr)r*r6rPr{is r rzSENet._make_layers  / Q;$--6EOO+CC MM6EOO#;QI_!+=EQMOQv7>2> J vvy&*[XZ[\0 q&! QA MM% vvyOBO P Q}}f%%r"c(td|rdnd}|S)Nz^layer0z ^layer(\d+)z^layer(\d+)\.(\d+))stemr)r)r5coarsematchers r group_matcherzSENet.group_matchersJ~Mbcr"c|rJdy)Nz$gradient checkpointing not supportedrV)r5enables r set_grad_checkpointingzSENet.set_grad_checkpointingsAAAz6r"returnc|jSrI)r)r5s r get_classifierzSENet.get_classifiersr"cp||_t|j|j|\|_|_y)N)r)rr rrr)r5rrs r reset_classifierzSENet.reset_classifiers3&->   t//;.H*$*r"c|j|}|j|}|j|}|j|}|j |}|j |}|SrI)rrrrrrr5r=s r forward_featureszSENet.forward_featuressU KKN JJqM KKN KKN KKN KKNr" pre_logitsc|j|}|jdkDr,tj||j|j}|r|S|j |S)Nr)ptraining)rr|Fdropoutrr)r5r=rs r forward_headzSENet.forward_headsP   Q  >>B  !t~~ FAq7D$4$4Q$77r"cJ|j|}|j|}|SrI)rrrs r r?z SENet.forwards'  ! !! $   a r") g?r:rqFr ravgNN)r r rNNF)T)r)r@rArBrrrgrrCfloatboolstrr/rtorchjitignorerrrrrrr?rDrEs@r rrs^ ##*+&'#$P ?P#s(OP P  P  PPPP%(P!$PPPdLMW[&& YY YYBB YY    HCHcH 8$8 r"c &tt||fi|SrI)r r)variant pretrainedkwargss r _create_senetrs w Ef EEr"c 2|dddddttdddd |S) Nr)r:r)rrg?bilinearz layer0.conv1rz apache-2.0) urlr input_size pool_sizecrop_pct interpolationr<std first_conv classifierlicenser)rrs r _cfgrs24}SYJ%.B$Ml     r"zmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/legacy_senet154-e9eb9fe6.pth)rzhhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/seresnet18-4bb0ce65.pthbicubic)rrzhhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/seresnet34-a4004e63.pthzhhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-cadene/se_resnet50-ce0d4300.pthzihttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-cadene/se_resnet101-7e38fcc6.pthzihttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-cadene/se_resnet152-d17c99b7.pthzphttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/seresnext26_32x4d-65ebdb501.pthzwhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/legacy_se_resnext50_32x4d-f3651bad.pthzxhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/legacy_se_resnext101_32x4d-37725eac.pth) zlegacy_senet154.in1kzlegacy_seresnet18.in1kzlegacy_seresnet34.in1kzlegacy_seresnet50.in1kzlegacy_seresnet101.in1kzlegacy_seresnet152.in1kzlegacy_seresnext26_32x4d.in1kzlegacy_seresnext50_32x4d.in1kzlegacy_seresnext101_32x4d.in1krc Zttgddd}td|fit|fi|S)Nr9r9r9r9r rrzr{r\r&legacy_seresnet18rrvrrr model_argss r rr3LbJJ ,j WDr sK  #(( A)*< )& ryy (8%:%PB*B*BII*ZKBIIK\F% {}" v!# vx" vx# w y# w y%) ~&!&* F&G&* G'H'& 0XUXX XUXX XUXX YeYY YeYY V5VV_E__ _E__ `U``r"