o ϯi' @sddlmZmZddlmZddlmZmZddlm Z dZ e de de de d d Z d d d efde de de dededejf ddZ dde de dedejfddZ dde de dedejfddZ dde de dedejfddZ dde de dedejfddZdS))AnyCallableN)create_slowfastcreate_slowfast_with_roi_head)load_state_dict_from_urlz5https://dl.fbaipublicfiles.com/pytorchvideo/model_zooz/kinetics/SLOWFAST_8x8_R50.pythz$/ava/SLOWFAST_8x8_R50_DETECTION.pythz /kinetics/SLOWFAST_8x8_R101.pythz'/kinetics/SLOWFAST_16x8_R101_50_50.pyth) slowfast_r50slowfast_r50_detection slowfast_r101slowfast_16x8_r101_50_50FT pretrainedprogresscheckpoint_path model_builderkwargsreturncKs6|di|}|rt||dd}|d}|||S)Ncpu)r map_location model_state)rload_state_dict)r r rrrmodel checkpoint state_dictrrT/home/ubuntu/.local/lib/python3.10/site-packages/pytorchvideo/models/hub/slowfast.py _slowfasts rcKtd||tdddd|S)a SlowFast R50 model architecture [1] trained with an 8x8 setting on the Kinetics dataset. Model with pretrained weights has top1 accuracy of 76.4. [1] Christoph Feichtenhofer et al, "SlowFast Networks for Video Recognition" https://arxiv.org/pdf/1812.03982.pdf Args: pretrained (bool): If True, returns a model pre-trained on Kinetics dataset progress (bool): If True, displays a progress bar of the download to stderr kwargs: use these to modify any of the other model settings. All the options are defined in pytorchvideo/models/slowfast.py NOTE: to use the pretrained model, do not modify the model configuration via the kwargs. Only modify settings via kwargs to initialize a new model without pretrained weights. r2)rr r r model_depth slowfast_fusion_conv_kernel_sizeNrrcheckpoint_pathsr r rrrrr%rcKr)a SlowFast R101 model architecture [1] trained with an 8x8 setting on the Kinetics dataset. Model with pretrained weights has top1 accuracy of 77.9. [1] Christoph Feichtenhofer et al, "SlowFast Networks for Video Recognition" https://arxiv.org/pdf/1812.03982.pdf Args: pretrained (bool): If True, returns a model pre-trained on Kinetics dataset progress (bool): If True, displays a progress bar of the download to stderr kwargs: use these to modify any of the other model settings. All the options are defined in pytorchvideo/models/slowfast.py NOTE: to use the pretrained model, do not modify the model configuration via the kwargs. Only modify settings via kwargs to initialize a new model without pretrained weights. r errr Nrr#r%rrrr Er&r c Ks&d}td||tddd|dd|S) a% SlowFast R101_50_50 model architecture [1] trained with an 16x8 setting on the Kinetics dataset. Model with pretrained weights has top1 accuracy of 78.7. [1] Christoph Feichtenhofer et al, "SlowFast Networks for Video Recognition" https://arxiv.org/pdf/1812.03982.pdf Args: pretrained (bool): If True, returns a model pre-trained on Kinetics dataset progress (bool): If True, displays a progress bar of the download to stderr kwargs: use these to modify any of the other model settings. All the options are defined in pytorchvideo/models/slowfast.py NOTE: to use the pretrained model, do not modify the model configuration via the kwargs. Only modify settings via kwargs to initialize a new model without pretrained weights. ))rrrr*rrr,r,r,r,r,r*r*r*r*r*r*r*r*r*r*r*r*r*r*r*r*r*r,)r,r,r+r,r r'r())rr)@rr)r r rr!r"stage_conv_a_kernel_sizeshead_pool_kernel_sizesNrr#)r r rr0rrrr esr cKstd||tdtd|S)aY SlowFast R50 model architecture [1] with pretrained weights based on 8x8 setting. The model is initially trained on Kinetics dataset for classification and later finetuned on AVA dataset for detection. [1] Christoph Feichtenhofer et al, "SlowFast Networks for Video Recognition" https://arxiv.org/pdf/1812.03982.pdf Args: pretrained (bool): If True, returns a model pre-trained on Kinetics dataset progress (bool): If True, displays a progress bar of the download to stderr kwargs: use these to modify any of the other model settings. All the options are defined in pytorchvideo/models/slowfast.py NOTE: to use the pretrained model, do not modify the model configuration via the kwargs. Only modify settings via kwargs to initialize a new model without pretrained weights. r)r r rrNr)rr$rr%rrrrsr)FT)typingrrtorch.nnnnpytorchvideo.models.slowfastrr torch.hubrroot_dirr$boolstrModulerrr r rrrrrs     ! ! 2