a  þdº@ã@s ddlZddlmZddlmmZddlZddlZddlm Z ddl m Z ddl m Z ddlmZddlmZe ¡Gdd„dejƒƒZGd d „d eƒZdS) éN)Ú flow_warp)ÚConvResidualBlocks)ÚSpyNet)ÚModulatedDeformConvPack)Ú ARCH_REGISTRYcsJeZdZdZd‡fdd „ Zd d „Zd d „Zdd„Zdd„Zdd„Z ‡Z S)ÚBasicVSRPlusPlusaBasicVSR++ network structure. Support either x4 upsampling or same size output. Since DCN is used in this model, it can only be used with CUDA enabled. If CUDA is not enabled, feature alignment will be skipped. Besides, we adopt the official DCN implementation and the version of torch need to be higher than 1.9. Paper: BasicVSR++: Improving Video Super-Resolution with Enhanced Propagation and Alignment Args: mid_channels (int, optional): Channel number of the intermediate features. Default: 64. num_blocks (int, optional): The number of residual blocks in each propagation branch. Default: 7. max_residue_magnitude (int): The maximum magnitude of the offset residue (Eq. 6 in paper). Default: 10. is_low_res_input (bool, optional): Whether the input is low-resolution or not. If False, the output resolution is equal to the input resolution. Default: True. spynet_path (str): Path to the pretrained weights of SPyNet. Default: None. cpu_cache_length (int, optional): When the length of sequence is larger than this value, the intermediate features are sent to CPU. This saves GPU memory, but slows down the inference speed. You can increase this number if you have a GPU with large memory. Default: 100. é@éé TNédc sÄtƒ ¡||_||_||_t|ƒ|_|r:td|dƒ|_nLt   t   d|ddd¡t j dddt   ||ddd¡t j dddt||dƒ¡|_t   ¡|_t   ¡|_gd¢}t|ƒD]J\}} tj ¡rÚtd||ddd |d |j| <td||||ƒ|j| <qªtd||dƒ|_t j ||d ddddd |_t j |d ddddd |_t  d¡|_t   ddddd¡|_t   ddddd¡|_t jd ddd|_t j ddd|_d|_t |jƒdkr°d|_!nd|_!t" #d¡dS)Nééééçš™™™™™¹?T©Znegative_slopeZinplace)Z backward_1Z forward_1Z backward_2Z forward_2é)ÚpaddingÚdeformable_groupsÚmax_residue_magnitudeé)ÚbiasérZbilinearF)Ú scale_factorÚmodeZ align_cornersrz›Deformable alignment module is not added. Probably your CUDA is not configured correctly. DCN can only be used with CUDA enabled. Alignment is skipped now.)$ÚsuperÚ__init__Ú mid_channelsÚis_low_res_inputÚcpu_cache_lengthrÚspynetrÚ feat_extractÚnnÚ SequentialÚConv2dÚ LeakyReLUZ ModuleDictÚ deform_alignÚbackboneÚ enumerateÚtorchÚcudaZ is_availableÚSecondOrderDeformableAlignmentÚreconstructionÚupconv1Úupconv2Z PixelShuffleÚ pixel_shuffleÚconv_hrÚ conv_lastZUpsampleÚ img_upsampleÚlreluÚis_mirror_extendedÚlenÚis_with_alignmentÚwarningsÚwarn) ÚselfrZ num_blocksrrZ spynet_pathrÚmodulesÚiÚmodule©Ú __class__©úf/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/basicsr/archs/basicvsrpp_arch.pyr*sN   ý   ú  zBasicVSRPlusPlus.__init__cCsH| d¡ddkrDtj|ddd\}}t || d¡¡dkrDd|_dS)aCheck whether the input is a mirror-extended sequence. If mirror-extended, the i-th (i=0, ..., t-1) frame is equal to the (t-1-i)-th frame. Args: lqs (tensor): Input low quality (LQ) sequence with shape (n, t, c, h, w). rrr©ZdimTN)Úsizer)ÚchunkZnormÚflipr4)r9ÚlqsÚlqs_1Úlqs_2r?r?r@Úcheck_if_mirror_extendedls z)BasicVSRPlusPlus.check_if_mirror_extendedc Csâ| ¡\}}}}}|dd…dd…dd…dd…dd…f d|||¡}|dd…dd…dd…dd…dd…f d|||¡}| ||¡ ||dd||¡} |jr¦|  d¡} n| ||¡ ||dd||¡} |jrÚ|  ¡} |  ¡} | | fS)auCompute optical flow using SPyNet for feature alignment. Note that if the input is an mirror-extended sequence, 'flows_forward' is not needed, since it is equal to 'flows_backward.flip(1)'. Args: lqs (tensor): Input low quality (LQ) sequence with shape (n, t, c, h, w). Return: tuple(Tensor): Optical flow. 'flows_forward' corresponds to the flows used for forward-time propagation (current to previous). 'flows_backward' corresponds to the flows used for backward-time propagation (current to next). Néÿÿÿÿrr)rBZreshaper Úviewr4rDÚ cpu_cacheÚcpu) r9rEÚnÚtÚcÚhÚwrFrGÚflows_backwardÚ flows_forwardr?r?r@Ú compute_flowzs22 zBasicVSRPlusPlus.compute_flowc s®| ¡\}}}}}td|dƒ} td|ƒ} ttdtˆdƒƒƒ} | | ddd…7} dˆvrl| ddd…} | } | ||j||¡} t| ƒD]\} ‰ˆd| ˆ}|jr¶| ¡}|  ¡} | dkrú|j rú|dd…| | dd…dd…dd…f}|jrü| ¡}t | |  dddd¡ƒ}t   | ¡}t   |¡}t   |¡}| dkrÀˆˆd }|jrV| ¡}|dd…| | ddd…dd…dd…f}|jr| ¡}|t ||  dddd¡ƒ}t ||  dddd¡ƒ}t j|||gdd }t j| |gdd } |jˆ| |||ƒ} |g‡‡‡fd d „ˆDƒ| g}|jr2d d „|Dƒ}t j|dd }| |jˆ|ƒ} ˆˆ | ¡|jr†ˆˆd ¡ˆˆd<t j ¡q†dˆvrªˆˆddd…ˆˆ<ˆS)a³Propagate the latent features throughout the sequence. Args: feats dict(list[tensor]): Features from previous branches. Each component is a list of tensors with shape (n, c, h, w). flows (tensor): Optical flows with shape (n, t - 1, 2, h, w). module_name (str): The name of the propgation branches. Can either be 'backward_1', 'forward_1', 'backward_2', 'forward_2'. Return: dict(list[tensor]): A dictionary containing all the propagated features. Each key in the dictionary corresponds to a propagation branch, which is represented by a list of tensors. rrrIÚspatialNÚbackwardrr éþÿÿÿrAcs$g|]}|dˆfvrˆ|ˆ‘qS)rUr?©Ú.0Úk©ÚfeatsÚidxÚ module_namer?r@Ú ×óz.BasicVSRPlusPlus.propagate..cSsg|] }| ¡‘qSr?)r*)rYÚfr?r?r@r_Ùr`)rBÚrangeÚlistr5Z new_zerosrr(rKr*r6rZpermuter)Z zeros_likeÚcatr&r'ÚappendrLÚ empty_cache)r9r\Úflowsr^rMrNÚ_rPrQZ frame_idxZflow_idxÚ mapping_idxZ feat_propr;Z feat_currentZflow_n1Zcond_n1Zfeat_n2Zflow_n2Zcond_n2ZcondÚfeatr?r[r@Ú propagate™s\ &     *"  zBasicVSRPlusPlus.propagatec slg}tˆdƒ}ttd|ƒƒ}||ddd…7}td| d¡ƒD]}‡fdd„ˆDƒ}| dˆd||¡tj|dd}|jrŒ| ¡}|  |¡}|  |  |  |¡¡¡}|  |  |  |¡¡¡}|  | |¡¡}| |¡}|jr|| |dd…|dd…dd…dd…f¡7}n&||dd…|dd…dd…dd…f7}|jrR| ¡}tj ¡| |¡q@tj|ddS) aDCompute the output image given the features. Args: lqs (tensor): Input low quality (LQ) sequence with shape (n, t, c, h, w). feats (dict): The features from the propgation branches. Returns: Tensor: Output HR sequence with shape (n, t, c, 4h, 4w). rUrNrIrcs"g|]}|dkrˆ| d¡‘qS)rUr)ÚpoprX©r\r?r@r_ùr`z-BasicVSRPlusPlus.upsample..rA)r5rcrbrBÚinsertr)rdrKr*r,r3r/r-r.r0r1rr2rLrfreÚstack)r9rEr\ÚoutputsZ num_outputsrir;Úhrr?rmr@Úupsampleès.    .&  zBasicVSRPlusPlus.upsamplec sú| ¡\}}}}}||jkr dnd|_|jr6| ¡}n2tj| d|||¡ddd ||||d|d¡}| |¡i}|jrÚg|d<t d |ƒD]H} |  |d d …| d d …d d …d d …f¡  ¡} |d  | ¡t j ¡qŽnV|  | d|||¡¡‰ˆjd d …\}}ˆ ||d||¡‰‡fd d „t d |ƒDƒ|d<| d¡dkrP| d¡dksfJd|›d|›dƒ‚| |¡\} } dD]t} dD]h}|›d| ›}g||<|dkrª| }n| d urº| }n |  d¡}| |||¡}|jr€~t j ¡q€qx| ||¡S)zóForward function for BasicVSR++. Args: lqs (tensor): Input low quality (LQ) sequence with shape (n, t, c, h, w). Returns: Tensor: Output HR sequence with shape (n, t, c, 4h, 4w). TFrIgÐ?Zbicubic)rrrrUrNrc s.g|]&}ˆdd…|dd…dd…dd…f‘qS)Nr?)rYr;©Zfeats_r?r@r_4r`z,BasicVSRPlusPlus.forward..r rzDThe height and width of low-res inputs must be at least 64, but got z and Ú.)rr)rVÚforwardrhrVr)rBrrKrÚcloneÚFZ interpolaterJrHrbr!rLrer)r*rfÚshaperTrDrkrr)r9rErMrNrOrPrQZlqs_downsampler\r;rjrSrRZiter_Ú directionr<rgr?rsr@rusV  ÿÿ ,"ÿÿÿ   zBasicVSRPlusPlus.forward)rr r TNr ) Ú__name__Ú __module__Ú __qualname__Ú__doc__rrHrTrkrrruÚ __classcell__r?r?r=r@rsúBO)rcs0eZdZdZ‡fdd„Zdd„Zdd„Z‡ZS)r+aÍSecond-order deformable alignment module. Args: in_channels (int): Same as nn.Conv2d. out_channels (int): Same as nn.Conv2d. kernel_size (int or tuple[int]): Same as nn.Conv2d. stride (int or tuple[int]): Same as nn.Conv2d. padding (int or tuple[int]): Same as nn.Conv2d. dilation (int or tuple[int]): Same as nn.Conv2d. groups (int): Same as nn.Conv2d. bias (bool or str): If specified as `auto`, it will be decided by the norm_cfg. Bias will be set as True if norm_cfg is None, otherwise False. max_residue_magnitude (int): The maximum magnitude of the offset residue (Eq. 6 in paper). Default: 10. csº| dd¡|_tt|ƒj|i|¤Žt t d|jd|jddd¡tj dddt |j|jddd¡tj dddt |j|jddd¡tj dddt |jd |j ddd¡¡|_ |  ¡dS) Nrr r rrrTré) rlrrr+rr"r#r$Z out_channelsr%rÚ conv_offsetÚ init_offset)r9ÚargsÚkwargsr=r?r@rcs   ù z'SecondOrderDeformableAlignment.__init__cCs"ddd„}||jdddddS)NrcSsLt|dƒr$|jdur$tj |j|¡t|dƒrH|jdurHtj |j|¡dS)NÚweightr)Úhasattrr„r"ÚinitZ constant_r)r<Úvalrr?r?r@Ú_constant_initvszBSecondOrderDeformableAlignment.init_offset.._constant_initrI)r‡r)r)r€)r9rˆr?r?r@rts z*SecondOrderDeformableAlignment.init_offsetc Csîtj|||gdd}| |¡}tj|ddd\}}}|jt tj||fdd¡} tj| ddd\} } | | d¡ d|  d¡ddd¡} | | d¡ d|  d¡ddd¡} tj| | gdd} t  |¡}t j   || |j |j|j|j|j|¡S)NrrAr r)r)rdr€rCrÚtanhrDÚrepeatrBZsigmoidÚ torchvisionÚopsZ deform_conv2dr„rZstriderZdilation) r9ÚxZ extra_featZflow_1Zflow_2ÚoutZo1Zo2ÚmaskÚoffsetZoffset_1Zoffset_2r?r?r@ru~s $$ ÿz&SecondOrderDeformableAlignment.forward)rzr{r|r}rrrur~r?r?r=r@r+Rs  r+)r)Ztorch.nnr"Ztorch.nn.functionalZ functionalrwr‹r7Zbasicsr.archs.arch_utilrZbasicsr.archs.basicvsr_archrZbasicsr.archs.spynet_archrZbasicsr.ops.dcnrZbasicsr.utils.registryrÚregisterÚModulerr+r?r?r?r@Ús      E