o oi@sddlmZmZmZmZmZmZmZddlZddl m Z m Z ddl m Z ddlmZmZmZedZGdddeZGd d d eed ZGd d d e ZGddde e ZdS))AnyCallableDictOptionalTupleTypeTypeVarN) DistributionUniform)MultiprocessWrapper)DeviceModuleTensorTc@s.eZdZdZdeedededefddZdS) _PostInitInjectionMetaClasszMTo inject the ``__post_init__`` function after the creation of each instance.clsargskwargsreturncOs$tj|g|Ri|}||SN)type__call__ __post_init__)rrrobjr]/home/ubuntu/.local/lib/python3.10/site-packages/kornia/augmentation/random_generator/base.pyr sz$_PostInitInjectionMetaClass.__call__N)__name__ __module__ __qualname____doc__rrrrrrrrrs"rc seZdZUdZdZeeed<ej ed<dfdd Z ddd Z dej fdeejdej ddfd d Z d ed eddfddZdejdej ddfddZddeedfdedeeeffddZZS)RandomGeneratorBasez9Base class for generating random augmentation parameters.NdevicedtypercstdSr)super__init__self __class__rrr$,szRandomGeneratorBase.__init__cCs |dSr)set_rng_device_and_dtyper%rrrr/s z!RandomGeneratorBase.__post_init__cCsF|dur td}|j|ks|j|kr!|||||_||_dSdS)zChange the random generation device and dtype. Note: The generated random numbers are not reproducible across different devices and dtypes. Ncpu)torchr!r" make_samplersr&r!r"rrrr)2s   z,RandomGeneratorBase.set_rng_device_and_dtyperrcOs.tjjj|i|\}}}}|j||d|S)N)r!r")r+_C_nn _parse_tor))r&rrr!r"_rrrtoEszRandomGeneratorBase.tocCtrNotImplementedErrorr-rrrr,Jz!RandomGeneratorBase.make_samplersF batch_shape. same_on_batchcCr3rr4)r&r7r8rrrforwardMr6zRandomGeneratorBase.forward)rN)F)rrrrr!rr __annotations__r+r"r$rfloat32r)rr2r,rintboolrstrrr9 __classcell__rrr'rr &s$    0r ) metaclassc@seZdZdZddedeeegefddfddZde e d fdefd d Z de e d fdefd d Z de defddZ dedefddZdS)DistributionWithMapperaWraps a distribution with a value mapper function. This is used to restrict the output values of a given distribution by a value mapper function. The value mapper function can be functions like sigmoid, tanh, etc. Args: dist: the target distribution. map_fn: the callable function to adjust the output from distributions. Example: >>> from torch.distributions import Normal >>> import torch.nn as nn >>> # without mapper >>> dist = DistributionWithMapper(Normal(0., 1.,), map_fn=None) >>> _ = torch.manual_seed(0) >>> dist.rsample((8,)) tensor([ 1.5410, -0.2934, -2.1788, 0.5684, -1.0845, -1.3986, 0.4033, 0.8380]) >>> # with sigmoid mapper >>> dist = DistributionWithMapper(Normal(0., 1.,), map_fn=nn.Sigmoid()) >>> _ = torch.manual_seed(0) >>> dist.rsample((8,)) tensor([0.8236, 0.4272, 0.1017, 0.6384, 0.2527, 0.1980, 0.5995, 0.6980]) Ndistmap_fnrcCs||_||_dSr)rBrC)r&rBrCrrrr$ks zDistributionWithMapper.__init__ sample_shape.cC*|jt|}|jdur||}|Sr)rBrsampler+SizerCr&rDoutrrrrFo  zDistributionWithMapper.rsamplecCrEr)rBsampler+rGrCrHrrrrKurJzDistributionWithMapper.samplencCs$|j|}|jdur||}|Sr)rBsample_nrC)r&rLrIrrrrM{s   zDistributionWithMapper.sample_nattrcCs,zt||WStyt|j|YSwr)getattrAttributeErrorrB)r&rNrrr __getattr__s   z"DistributionWithMapper.__getattr__r)rrrrr rrrr$rr<rFrKrMr>rrQrrrrrAQs&rAc@seZdZdZdS)UniformDistributionzgWrapper around torch Uniform distribution which makes it work with the 'spawn' multiprocessing context.N)rrrrrrrrrRsrR)typingrrrrrrrr+torch.distributionsr r !kornia.augmentation.utils.helpersr kornia.corer r rrrrr rArRrrrrs$  +7