3j+t>%SSKJr SSKrSSKJr SSKrSSKJs Jr SSKJr SSK J r J r SSK Jr SSKJrJrJr SSKJr SS KJr 0rS \S 'S S /rS"SjrS"SjrS#S$SjjrS%SjrS%SjrS"SjrS&Sjr S"Sjr!S"Sjr"S'Sjr#S(Sjr$S(Sjr%S)Sjr&S*Sjr'S+Sjr(S,Sjr)S-Sjr*\S.S/Sjj5r+S0S jr,"S!S \RZ5r.g)1) annotationsN)Union)nn) rgb_to_ycbcr ycbcr_to_rgb)pi) KORNIA_CHECKKORNIA_CHECK_IS_TENSORKORNIA_CHECK_SHAPErescale)perform_keep_shape_imagezhdict[tuple[Union[torch.dtype, None], Union[str, torch.device, None]], tuple[torch.Tensor, torch.Tensor]] _DCT8_CACHEJPEGCodecDifferentiablejpeg_codec_differentiablec6UR5nXU- S--$)zDifferentiable rounding. Args: input: Input tensor of any shape to be rounded. Returns: Pseudo rounded tensor of the same shape as input tensor. )round)input input_rounds M/home/wildlama/miniconda3/lib/python3.13/site-packages/kornia/enhance/jpeg.py#_differentiable_polynomial_roundingr,s"++-K +-!3 33c<UR5nXS- U- S--$)zPerform floor via a differentiable operation. Args: input: Input tensor of any shape to be floored. Returns: Pseudo rounded tensor of the same shape as input tensor. ?r)floor)r input_floors r _differentiable_polynomial_floorr:s&++-K #+ 39 99rcUR5nUb.U*[R"XDU:*U-5S- -U-XDU:'Ub,U[R"XDU:U- 5S- -U-XDU:'U$)a#Clip via a differentiable and soft approximation of the clipping operation. Args: input: Input tensor of any shape. min_val: Minimum value. max_val: Maximum value. scale: Scale value. Default 0.02. Returns: Clipped output tensor of the same shape as the input tensor. ?)clonetorchexp)rmin_valmax_valscaleoutputs r_differentiable_clippingr(Hs$!;;=F$)6UYY?O8P7PSZ7Z-[^a-a#bel#l #(EIIfg=M6NQX6X,Y\_,_#`cj#j Mrc ^[R"/SQ/SQ/SQ/SQ/SQ/SQ/SQ/SQ/UUS 9$) z1Generate default Quantization table of Y channel.) r*(3=) r1:<7)r2 r*r-r.9E8)r2r/WP>)r=%r;DmgM)r-#r7@Qhq\)1rINr?rFyxe)HrM_bpdrFcdevicedtyper"tensorrYs r_get_default_qt_yr^bs8 << , , , , . . 0 /   rc ^[R"/SQ/SQ/SQ/SQ/SQ/SQ/SQ/SQ/UUS9$)z2Generate default Quantization table of C channels.)r<rBr-/rXrXrXrX)rBr4BrXrXrXrX)r-r4r;rXrXrXrXrX)r`rbrXrXrXrXrXrX)rXrXrXrXrXrXrXrXrYr\rYs r_get_default_qt_crcts8 << , , , , , , , ,   rcURupnURXS-SUS-S5RSSSSS5RUSSS5nU$)zExtract non-overlapping 8 x 8 patches from the given input image. Args: input (torch.Tensor): Input image of the shape :math:`(B, H, W)`. Returns: output (torch.Tensor): Image patchify of the shape :math:`(B, N, 8, 8)`. rrshapeviewpermutereshape)rBHWr's r _patchify_8x8rrs\kkGA! ::aaAFA>FFq!QPQSTU]]^_acefhijF MrcURSSup4URX1S-US-SS5RSSSSS5RX1U5nU$)aReverse non-overlapping 8 x 8 patching. Args: input (torch.Tensor): Input image of the shape :math:`(B, N, 8, 8)`. H: height of resulting torch.Tensor. W: width of resulting torch.Tensor. Returns: output (torch.Tensor): Image patchify of the shape :math:`(B, H, W)`. Nrgrerrfrrhrj)rrprqro_Nr's r_unpatchify_8x8rusZ KKOEA ::aaaA>FFq!QPQSTU]]^_defF MrcURnURn[X5up4US[R"US- U5-nU$)zPerform an 8 x 8 discrete cosine transform. Args: input (torch.Tensor): Patched input torch.tensor of the shape :math:`(B, N, 8, 8)`. Returns: output (torch.Tensor): DCT output torch.tensor of the shape :math:`(B, N, 8, 8)`. NN`@)r[rZ_get_dct8_basis_scaler" tensordot)rr[rZ dct_tensor dct_scaler's r_dct_8x8r}sG',kkE-2\\F1%@J$Z05??55=R\3]]F MrcURnURn[R"SXS9nUR S5nUR S5n[R "SU-S-U-[ -S- 5n[R"SXS9nSUS'[R"Xw5nX-nX-n U RS S 5U-RS S 5n U S -S -n U $) zPerform an 8 x 8 inverse discrete cosine transform. Args: input (torch.Tensor): Patched input torch.tensor of the shape :math:`(B, N, 8, 8)`. Returns: output (torch.Tensor): IDCT output torch.tensor of the shape :math:`(B, N, 8, 8)`. rer[rZrrf@r 0@;f?ri?rx) r[rZr"arange unsqueezecosronesouter transpose) rr[rZidx spatial_idxfreq_idxbasisalphar|tmpr's r _idct_8x8rs KKE \\F ,,q 5C--"K}}QH IIs[(3.(:R?$F GE JJq 5EE!H E)I  E -CmmB#e+ 6 6r2 >F d]U "F Mrc `[[R"US:SU- SSU-- 55nU$)aConvert a given JPEG quality to the scaling factor. Args: compression_strength (torch.Tensor): Compression strength ranging from 0 to 100. Any shape is supported. Returns: scale (torch.Tensor): Scaling factor to be applied to quantization matrix. Same shape as input. 2g@gi@r)rr"where)compression_strengthr&s r_jpeg_quality_to_scalers@; 2 % ) ) C.. . E LrcUSS2S4[U5SS2SSS4-n[[US-S- SS55nX- n[U5nU$)aPerform quantization. Args: input (torch.Tensor): Input torch.tensor of the shape :math:`(B, N, 8, 8)`. jpeg_quality (torch.Tensor): Compression strength to be applied, shape is :math:`(B)`. quantization_table (torch.Tensor): Quantization table of the shape :math:`(1, 8, 8)` or :math:`(B, 8, 8)`. Returns: output (torch.Tensor): Quantized output torch.tensor of the shape :math:`(B, N, 8, 8)`. NI@Y@rf)rrr(rr jpeg_qualityquantization_tablequantization_table_scaledr's r _quantizersn$ 1d7#&<\&J1dTXZ^K^&__: ";d"Be!KQPST!5F 0 8F MrcUSS2S4[U5SS2SSS4-nU[[US-S- SS55-nU$)aPerform dequantization. Args: input (torch.Tensor): Input torch.tensor of the shape :math:`(B, N, 8, 8)`. jpeg_quality (torch.Tensor): Compression strength to be applied, shape is :math:`(B)`. quantization_table (torch.Tensor): Quantization table of the shape :math:`(1, 8, 8)` or :math:`(B, 8, 8)`. Returns: output (torch.Tensor): Quantized output torch.tensor of the shape :math:`(B, N, 8, 8)`. Nrrrfr)rrr(rs r _dequantizersc$ 1d7#&<\&J1dTXZ^K^&__!#C ";d"Be!KQPST$F MrcUSS2S4nUSS2S4nUSS2S4n[USS2S4SSSSS 9n[USS2S4SSSSS 9nXSS2S4USS2S44$) aImplement chroma subsampling. Args: input_ycbcr (torch.Tensor): YCbCr input torch.tensor of the shape :math:`(B, 3, H, W)`. Returns: output_y (torch.Tensor): Y component (not-subsampled), shape is :math:`(B, H, W)`. output_cb (torch.Tensor): Cb component (subsampled), shape is :math:`(B, H // 2, W // 2)`. output_cr (torch.Tensor): Cr component (subsampled), shape is :math:`(B, H // 2, W // 2)`. NrrfrgrbilinearFTfactor interpolation align_corners antialiasr ) input_ycbcroutput_y output_cb output_crs r_chroma_subsamplingr+s)A.H)!Q$/I)!Q$/I!T'  I!T'  I q!t_i1o 55rc<[USS2S4SSSSS9nUSS2S4$)zPerform chroma upsampling. Args: input_c (torch.Tensor): Cb or Cr component to be upsampled of the shape :math:`(B, H, W)`. Returns: output_c (torch.Tensor): Upsampled C(b or r) component of the shape :math:`(B, H * 2, W * 2)`. NrrFrrr )input_coutput_cs r_chroma_upsamplingrMs6%4  H AqD>rc.[U5nSU-n[U5upVn[U5[U5[U5pvn[U5n[[R "Xg4SS95n [ UUU5n [ U UU5RSSS9upXU 4$)aePerform JPEG encoding. Args: image_rgb (torch.Tensor): RGB input images of the shape :math:`(B, 3, H, W)`. jpeg_quality (torch.Tensor): Compression strength of the shape :math:`(B)`. quantization_table_y (torch.Tensor): Quantization table for Y channel. quantization_table_c (torch.Tensor): Quantization table for C channels. Returns: y_encoded (torch.Tensor): Encoded Y component of the shape :math:`(B, N, 8, 8)`. cb_encoded (torch.Tensor): Encoded Cb component of the shape :math:`(B, N, 8, 8)`. cr_encoded (torch.Tensor): Encoded Cr component of the shape :math:`(B, N, 8, 8)`. o@rfdimrg)rrrrr}r"catrchunk) image_rgbrquantization_table_yquantization_table_c image_ycbcrinput_yinput_cbinput_crdct_y dct_cb_cr y_encoded cb_encoded cr_encodeds r _jpeg_encoderbs*!-Y 7K+%K"5k"BGx ghh G W EH#7Q?@I' I ' eA1eo J * ,,rc[UUU5n[[R"X4SS9UU5n[U5n [U5R SSS9up[ XU5n [ XS-US-5n [ XS-US-5n[ U 5n [ U5n[R"XU4SS9S- n[U5nU$)aPerform JPEG decoding. Args: input_y (torch.Tensor): Compressed Y component of the shape :math:`(B, N, 8, 8)`. input_cb (torch.Tensor): Compressed Cb component of the shape :math:`(B, N, 8, 8)`. input_cr (torch.Tensor): Compressed Cr component of the shape :math:`(B, N, 8, 8)`. jpeg_quality (torch.Tensor): Compression strength of the shape :math:`(B)`. H (int): Original image height. W (int): Original image width. quantization_table_y (torch.Tensor): Quantization table for Y channel. quantization_table_c (torch.Tensor): Quantization table for C channels. Returns: rgb_decoded (torch.Tensor): Decompressed RGB image of the shape :math:`(B, 3, H, W)`. rfrrgr) rr"rrrrurstackr)rrrrrprqrr input_cb_cridct_yidct_cbidct_crimage_yimage_cbimage_crr rgb_decodeds r _jpeg_decoders6G  8&A.K %W-F -33A13=G+Fq9G,W1fa1fEH,W1fa1fEH!(+H!(+H % W,Iq QTY YK ,[ 9K rcURSSup[R"US- 5S-U- n[R"US- 5S-U- n[R"USUSU4S5nXSU4$)a#Pad a given image to be dividable by 16. Args: image: Image of the shape :math:`(*, 3, H, W)`. Returns: image_padded: Padded image of the shape :math:`(*, 3, H_{new}, W_{new})`. h_pad: Padded pixels along the horizontal axis. w_pad: Padded pixels along the vertical axis. rNr*r replicate)rkmathceilFpad)imagerprqh_padw_pad image_paddeds r_perform_paddingrst ;;rs DA1r6"R'!+E1r6"R'!+E!"uq%E.BK!PL  %%rc "[U5 [U5 URnURnUc [XT5OUnUc [ XT5OUn[U5 [U5 [ U/SQ5 [ US/5 UR S:XaURSS9nUR S:XaURSS9n[ U/SQ5 [ U/SQ5 [UR5R5S:=(a! UR5R5S :*S UR5R5S UR5R5S 35 [U5upnURS SupURSS:waJ[URSURS:HSURSSURSS35 URSS:waJ[URSURS:HSURSSURSS35 URSS:waJ[URSURS:HSURSSURSS35 URXT5nURXT5nURXT5n[UUUUS9upn [!U U U UUU UUS9n [#U SSS9n U SSX- 2SX- 24n U $)a8Differentiable JPEG encoding-decoding module. Based on :cite:`reich2024` :cite:`shin2017`, we perform differentiable JPEG encoding-decoding as follows: .. image:: _static/img/jpeg_codec_differentiable.png .. math:: \text{JPEG}_{\text{diff}}(I, q, QT_{y}, QT_{c}) = \hat{I} Where: - :math:`I` is the original image to be coded. - :math:`q` is the JPEG quality controlling the compression strength. - :math:`QT_{y}` is the luma quantization table. - :math:`QT_{c}` is the chroma quantization table. - :math:`\hat{I}` is the resulting JPEG encoded-decoded image. .. note::: The input (and output) pixel range is :math:`[0, 1]`. In case you want to handle normalized images you are required to first perform denormalization followed by normalizing the output images again. Note, that this implementation models the encoding-decoding mapping of JPEG in a differentiable setting, however, does not allow the excess of the JPEG-coded byte file itself. For more details please refer to :cite:`reich2024`. This implementation is not meant for data loading. For loading JPEG images please refer to `kornia.io`. There we provide an optimized Rust implementation for fast JPEG loading. Args: image_rgb: the RGB image to be coded. jpeg_quality: JPEG quality in the range :math:`[0, 100]` controlling the compression strength. quantization_table_y: quantization table for Y channel. Default: `None`, which will load the standard quantization table. quantization_table_c: quantization table for C channels. Default: `None`, which will load the standard quantization table. Shape: - image_rgb: :math:`(*, 3, H, W)`. - jpeg_quality: :math:`(1)` or :math:`(B)` (if used batch dim. needs to match w/ image_rgb). - quantization_table_y: :math:`(8, 8)` or :math:`(B, 8, 8)` (if used batch dim. needs to match w/ image_rgb). - quantization_table_c: :math:`(8, 8)` or :math:`(B, 8, 8)` (if used batch dim. needs to match w/ image_rgb). Return: JPEG coded image of the shape :math:`(B, 3, H, W)` Example: To perform JPEG coding with the standard quantization tables just provide a JPEG quality >>> img = torch.rand(3, 3, 64, 64, requires_grad=True, dtype=torch.float) >>> jpeg_quality = torch.tensor((99.0, 25.0, 1.0), requires_grad=True) >>> img_jpeg = jpeg_codec_differentiable(img, jpeg_quality) >>> img_jpeg.sum().backward() You also have the option to provide custom quantization tables >>> img = torch.rand(3, 3, 64, 64, requires_grad=True, dtype=torch.float) >>> jpeg_quality = torch.tensor((99.0, 25.0, 1.0), requires_grad=True) >>> quantization_table_y = torch.randint(1, 256, size=(3, 8, 8), dtype=torch.float) >>> quantization_table_c = torch.randint(1, 256, size=(3, 8, 8), dtype=torch.float) >>> img_jpeg = jpeg_codec_differentiable(img, jpeg_quality, quantization_table_y, quantization_table_c) >>> img_jpeg.sum().backward() In case you want to control the quantization purly base on the quantization tables use a JPEG quality of 99.5. Setting the JPEG quality to 99.5 leads to a QT scaling of 1, see Eq. 2 of :cite:`reich2024` for details. >>> img = torch.rand(3, 3, 64, 64, requires_grad=True, dtype=torch.float) >>> jpeg_quality = torch.ones(3) * 99.5 >>> quantization_table_y = torch.randint(1, 256, size=(3, 8, 8), dtype=torch.float) >>> quantization_table_c = torch.randint(1, 256, size=(3, 8, 8), dtype=torch.float) >>> img_jpeg = jpeg_codec_differentiable(img, jpeg_quality, quantization_table_y, quantization_table_c) >>> img_jpeg.sum().backward() N)*3rprqrorgrr)ro8rgrz?JPEG quality is out of range. Expected range is [0, 100], got [z, z"]. Consider clipping jpeg_quality.rrfz#Batch dimensions do not match. Got z images and z quantization tables (Y).z quantization tables (C).z JPEG qualities.)rrrr)rrrrrprqrrr)rr$r%.)r r[rZr^rcr ndimrr aminitemamaxrrktorrr()rrrrr[rZrrrprqrrrimage_rgb_jpegs rrrsQb9%<(&/ooE-6-=-=F?S?[,V;au?S?[,V;au/0/0y"67|cU+  A%3==!=D  A%3==!=D+_=+_=     ! ! #s *U1B1B1D1I1I1Ku1T !!#((*+2l.?.?.A.F.F.H-IIk m /y9Ie ??23 DA!!!$) & &q )Y__Q-? ???1%&l3G3M3Ma3P2QQj l !!!$) & &q )Y__Q-? ???1%&l3G3M3Ma3P2QQj l !!   q !Y__Q%7 7??1%&l<3E3Ea3H2IIY [  ??61L/226A/226A(4!11 )%I: $0! 11 $N.NCY^_N#C19k k$ABN rczX4nU[;a[R"SXS9nSU-S-SS2S4USSS24-[S- -n[R"U5nUSS2SSS2S4USSS2SSS24-n[R "SXS9nSUS'[R "Xw5S-nXh4[U'[U$) Nrerrr rrrr)rr"rrrrr) r[rZkeyifreqbasis_1dr{rr|s rryry}s /C + LL% 7a# q$w'!D!G*4T B99T?aq$./(4D!;K2LL  1E9!aKK-4 &2 C s rcX^\rSrSrSrSSU4SjjjrSSjrSrU=r$) ria Differentiable JPEG encoding-decoding module. Based on :cite:`reich2024` :cite:`shin2017`, we perform differentiable JPEG encoding-decoding as follows: .. math:: \text{JPEG}_{\text{diff}}(I, q, QT_{y}, QT_{c}) = \hat{I} Where: - :math:`I` is the original image to be coded. - :math:`q` is the JPEG quality controlling the compression strength. - :math:`QT_{y}` is the luma quantization table. - :math:`QT_{c}` is the chroma quantization table. - :math:`\hat{I}` is the resulting JPEG encoded-decoded image. .. image:: _static/img/jpeg_codec_differentiable.png .. note:: The input (and output) pixel range is :math:`[0, 1]`. In case you want to handle normalized images you are required to first perform denormalization followed by normalizing the output images again. Note, that this implementation models the encoding-decoding mapping of JPEG in a differentiable setting, however, does not allow the excess of the JPEG-coded byte file itself. For more details please refer to :cite:`reich2024`. This implementation is not meant for data loading. For loading JPEG images please refer to `kornia.io`. There we provide an optimized Rust implementation for fast JPEG loading. Args: quantization_table_y: quantization table for Y channel. Default: `None`, which will load the standard quantization table. quantization_table_c: quantization table for C channels. Default: `None`, which will load the standard quantization table. Shape: - quantization_table_y: :math:`(8, 8)` or :math:`(B, 8, 8)` (if used batch dim. needs to match w/ image_rgb). - quantization_table_c: :math:`(8, 8)` or :math:`(B, 8, 8)` (if used batch dim. needs to match w/ image_rgb). - image_rgb: :math:`(*, 3, H, W)`. - jpeg_quality: :math:`(1)` or :math:`(B)` (if used batch dim. needs to match w/ image_rgb). Example: You can use the differentiable JPEG module with standard quantization tables by >>> diff_jpeg_module = JPEGCodecDifferentiable() >>> img = torch.rand(2, 3, 32, 32, requires_grad=True, dtype=torch.float) >>> jpeg_quality = torch.tensor((99.0, 1.0), requires_grad=True) >>> img_jpeg = diff_jpeg_module(img, jpeg_quality) >>> img_jpeg.sum().backward() You can also specify custom quantization tables to be used by >>> quantization_table_y = torch.randint(1, 256, size=(2, 8, 8), dtype=torch.float) >>> quantization_table_c = torch.randint(1, 256, size=(2, 8, 8), dtype=torch.float) >>> diff_jpeg_module = JPEGCodecDifferentiable(quantization_table_y, quantization_table_c) >>> img = torch.rand(2, 3, 32, 32, requires_grad=True, dtype=torch.float) >>> jpeg_quality = torch.tensor((99.0, 1.0), requires_grad=True) >>> img_jpeg = diff_jpeg_module(img, jpeg_quality) >>> img_jpeg.sum().backward() In case you want to learn the quantization tables just pass parameters `nn.Parameter` >>> quantization_table_y = torch.nn.Parameter(torch.randint(1, 256, size=(2, 8, 8), dtype=torch.float)) >>> quantization_table_c = torch.nn.Parameter(torch.randint(1, 256, size=(2, 8, 8), dtype=torch.float)) >>> diff_jpeg_module = JPEGCodecDifferentiable(quantization_table_y, quantization_table_c) >>> img = torch.rand(2, 3, 32, 32, requires_grad=True, dtype=torch.float) >>> jpeg_quality = torch.tensor((99.0, 1.0), requires_grad=True) >>> img_jpeg = diff_jpeg_module(img, jpeg_quality) >>> img_jpeg.sum().backward() cv>[TU]5 Uc [SS5OUnUc [SS5OUn[ U[ R 5(aURSU5 OURSU5 [ U[ R 5(aURSU5 gURSU5 g)Nrr) super__init__r^rc isinstancer Parameterregister_parameterregister_buffer)selfrr __class__s rr JPEGCodecDifferentiable.__init__s @T@\0trsC$#  3 67  %&A B 4 : #'"&        4$$"&&<&. % < % 66D*,-,-,-',-' ,- 5 ,-^3 333 3  3  3'3'33l&,1504 \\\.\. \  \\~  # -K &  lbiilr