U ʹjl@sdZddlmZmZmZmZmZddlZddlmZddl m Z m Z zddl m ZdZWnek rpd ZYnXdd lmZmZmZmZdd lmZdd lmZmZdd lmZmZmZmZm Z m!Z!m"Z"dddddddddg Z#eeeeeeeeeeee e!e"hZ$zddlm%Z%e$&e%Wnek r0YnXeej'dddZ(eej'dddZ)ddZ*e+Z,edddZ-edddZ.ddZ/ej'eee0e0fee0fd d!dZ Gd"ddej'Z1Gd#ddej'Z2dS)$zm PyTorch FX Based Feature Extraction Helpers Using https://pytorch.org/vision/stable/feature_extraction.html )CallableListDictUnionTypeN)nn)_get_feature_info_get_return_layers)create_feature_extractorTF) Conv2dSameScaledStdConv2dSame CondConv2d StdConv2dSame)BilinearAttnTransform) MaxPool2dSame AvgPool2dSame)BatchNormAct2dSyncBatchNormActFrozenBatchNormAct2d GroupNormAct GroupNorm1Act LayerNormActLayerNormAct2dregister_notrace_moduleis_notrace_moduleget_notrace_modulesregister_notrace_functionis_notrace_functionget_notrace_functionsr FeatureGraphNetGraphExtractNet) InplaceAbnmodulecCst||S)zq Any module not under timm.models.layers should get this decorator if we don't want to trace through it. ) _leaf_modulesaddr#r'W/var/www/html/bwave/python/venv/lib/python3.8/site-packages/timm/models/_features_fx.pyr;s cCs|tkSN)r%r#r'r'r(rCscCsttSr))listr%r'r'r'r(rGsfunccCst||S)zF Decorator for functions which ought not to be traced through )_autowrap_functionsr&r+r'r'r(rOs cCs|tkSr))r-r+r'r'r(rWscCsttSr))r*r-r'r'r'r(r[smodel return_nodescCs(ts tdt||ttttddS)NKPlease update to PyTorch 1.10+, torchvision 0.11+ for FX feature extraction)Z leaf_modulesZautowrap_functions)Z tracer_kwargs)has_fx_feature_extractionAssertionError_create_feature_extractorr*r%r-r.r'r'r(r _s  cs*eZdZdZdfdd ZddZZS)r zX A FX Graph based feature extractor that works with the model feature_info metadata NcsZttstdt|||_|dk r>t|t|ks>tt|j|}t|||_ dS)Nr1) super__init__r2r3r Z feature_infolenr r graph_module)selfr/Z out_indicesZout_mapr0 __class__r'r(r6js    zFeatureGraphNet.__init__cCst||Sr))r*r8values)r9xr'r'r(forwardsszFeatureGraphNet.forward)N)__name__ __module__ __qualname____doc__r6r> __classcell__r'r'r:r(r gs cs\eZdZdZd eeeefeefedfdd Z eee j e j fdddZ Z S) r!a A standalone feature extraction wrapper that maps dict -> list or single tensor NOTE: * one can use feature_extractor directly if dictionary output is desired * unlike FeatureGraphNet, this is intended to be used standalone and not with model feature_info metadata for builtin feature extraction mode * create_feature_extractor can be used directly if dictionary output is acceptable Args: model: model to extract features from return_nodes: node names to return features from (dict or list) squeeze_out: if only one output, and output in list format, flatten to single tensor T)r0 squeeze_outcs t||_t|||_dSr))r5r6rDr r8)r9r/r0rDr:r'r(r6s zGraphExtractNet.__init__)returncCs0t||}|jr,t|dkr,|dS|S)Nrr)r*r8r<rDr7)r9r=outr'r'r(r>szGraphExtractNet.forward)T)r?r@rArBrrstrrboolr6torchZTensorr>rCr'r'r:r(r!ws *)3rBtypingrrrrrrIrZ _featuresr r Z%torchvision.models.feature_extractionr r4r2 ImportErrorZ timm.layersr r rrZtimm.layers.non_local_attnrZtimm.layers.pool2d_samerrZtimm.layers.norm_actrrrrrrr__all__r%r"r&Modulerrrsetr-rrrrGr r!r'r'r'r(sf    $  &