+ jLRt^RIHt^RIt^RIHt^RIHt^RIHt^RI H t ^RI H t H t Ht]P!]4tRR ltR R ltR R lt] !RR7]!RR] 444t!RR]P,4t!RR]P,4t!RR]P,4t!RR]P,4t] !RR]44t] !RR7!RR]44tRR.tR#)zPyTorch SuperPoint model.) dataclassN)nn)PreTrainedModel)BaseModelOutputWithNoAttentionSuperPointConfig) ModelOutputauto_docstringloggingcV^8dQhR\PR\PR\R\R\R\\P\P3,/#) keypointsscoresborderheightwidthreturntorchTensorinttuple)formats"s/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/models/superpoint/modeling_superpoint.py __annotate__r%sV))||)%*\\);>)HK)TW) 5<< %&)cVR,V8VR,W2, 8,pVR,V8VR,WB, 8,pWV,pW,W,3#)zPRemoves keypoints (and their associated scores) that are too close to the border)NNN)r)r rrrrmask_hmask_wmasks&&&&& rremove_keypoints_from_bordersr$%sVo'IdOv,O PFo'IdOu~,N OF ?D ?FL ((rc V^8dQhR\PR\PR\R\\P\P3,/#)r r rkrr)rs"rrr/sH&&u||&U\\&c&eTYT`T`bgbnbnTnNo&rcpV\V48dW3#\P!W^R7wrW,V3#)z(Keeps the k keypoints with highest score)dim)lenrtopk)r rr&indicess&&& rtop_k_keypointsr,/s7C N  jj2OF  v %%rcdV^8dQhR\PR\R\P/#)r r nms_radiusr)rrr)rs"rrr7s)00u||000rcZaS^8d \R4hV3Rlp\P!V4pW!V48Hp\^4FMpV!VP 44^8p\P !WcV4pWr!V48HpWHV(,,pKO \P !W@V4#)z)Applies non-maximum suppression on scoresz'Expected positive values for nms_radiuscf<\PPVS^,^,^SR7#)r  kernel_sizestridepadding)r functional max_pool2d)xr.s&rmax_poolsimple_nms..max_pool<s,}}''zA~7IRS]g'hhr) ValueErrorr zeros_likerangefloatwhere) rr.r8zerosmax_mask_ supp_mask supp_scores new_max_masks &f r simple_nmsrE7sA~BCCi   V $E&))H 1XX^^-.2 kk)F; "h{&;;  z:;  ;;x //ra Base class for outputs of image point description models. Due to the nature of keypoint detection, the number of keypoints is not fixed and can vary from image to image, which makes batching non-trivial. In the batch of images, the maximum number of keypoints is set as the dimension of the keypoints, scores and descriptors tensors. The mask tensor is used to indicate which values in the keypoints, scores and descriptors tensors are keypoint information and which are padding. ) custom_introcHa]tRt^ItoRtRtRtRtRtRt Rt V3Rlt Rt Vt R#)#SuperPointKeypointDescriptionOutputa8 loss (`torch.FloatTensor` of shape `(1,)`, *optional*): Loss computed during training. keypoints (`torch.FloatTensor` of shape `(batch_size, num_keypoints, 2)`): Relative (x, y) coordinates of predicted keypoints in a given image. scores (`torch.FloatTensor` of shape `(batch_size, num_keypoints)`): Scores of predicted keypoints. descriptors (`torch.FloatTensor` of shape `(batch_size, num_keypoints, descriptor_size)`): Descriptors of predicted keypoints. mask (`torch.BoolTensor` of shape `(batch_size, num_keypoints)`): Mask indicating which values in keypoints, scores and descriptors are keypoint information. hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, + one for the output of each stage) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states (also called feature maps) of the model at the output of each stage. Nc><V^8dQh/S[PR,;R&S[PR,;R&S[PR,;R&S[PR,;R&S[PR,;R&S[S[P,R,;R&#)r Nlossr r descriptorsr# hidden_states)r FloatTensor IntTensor BoolTensorr)r __classdict__s"rr0SuperPointKeypointDescriptionOutput.__annotate__Is:   d ");<%,=>    $+?@""T)0AB   T !(CD**+d29Err )__name__ __module__ __qualname____firstlineno____doc__rJr rrKr#rL__annotate_func____static_attributes____classdictcell__rPs@rrHrHIs2$&*D(,I'+F,0K$(D59MErrHcTaa]tRt^ntoRV3RlV3RllltV3RlRltRtVtV;t#)SuperPointConvBlockc 6<V^8dQhRS[RS[RS[RS[RR/#)r config in_channels out_channels add_poolingrN)rrbool)rrPs"rr SuperPointConvBlock.__annotate__os<SS&S58SHKSZ^S Src8<\SV`4\P!VV^^^R7Vn\P!VV^^^R7Vn\P !RR7VnV'd \P!^^R7Vn R#RVn R#)r1Tinplacer2r3N) super__init__rConv2dconv_aconv_bReLUrelu MaxPool2dpool)selfr^r_r`ra __class__s&&&&&rrjSuperPointConvBlock.__init__os| ii     ii     GGD) =HBLLQq9 d rcN<V^8dQhRS[PRS[P/#)r rLrrr)rrPs"rrrcs#U\\ellrcVPVPV44pVPVPV44pVPeVPV4pV#N)rorlrmrq)rrrLs&&rforwardSuperPointConvBlock.forwardsN $++m"<=  $++m"<= 99  IIm4Mr)rlrmrqro)F) rRrSrTrUrjryrXrY __classcell__rsrPs@@rr\r\ns SS*rr\cXaa]tRt^toRtV3RlV3RlltRV3RlRlltRtVtV;t #)SuperPointEncoderz SuperPoint encoder module. It is made of 4 convolutional layers with ReLU activation and max pooling, reducing the dimensionality of the image. c$<V^8dQhRS[RR/#r r^rNr)rrPs"rrSuperPointEncoder.__annotate__s66/6D6rc @<\SV`4^Vn.pVP\ WPVP ^,RR74\ ^\VP 4^, 4FIpVP\ WP V^, ,VP V,RR74KK VP\ WP R,VP R,RR74\P!V4Vn R#)rT)raFN) rirj input_dimappendr\encoder_hidden_sizesr<r)r ModuleList conv_blocks)rrr^rirss&& rrjSuperPointEncoder.__init__s   8S8STU8Vdh i q#f99:Q>?A   #77A>@[@[\]@^lp @  33B79T9TUW9Xfk  ==5rcX<V^8dQhRS[R,RS[R,RS[S[,/#)r output_hidden_statesN return_dictr)rbrr)rrPs"rrrs9  #Tk D[  / /  rcV'dRMRpVPFpV!V4pV'gKWA3,pK TpV'g/\;QJd.RWd34FNK 5#!RWd344#\VVR7#)Nc30"TF qfKVxK R#5irxr .0vs& r ,SuperPointEncoder.forward..sQ$?q$? )last_hidden_staterLr )rrr)rrinputrrall_hidden_states conv_blockoutputs&&&& rrySuperPointEncoder.forwards~ #7BD**Ju%E##$5$@!+5QV$?Q5 Q5QV$?QQ Q-$+  r)rr)FT) rRrSrTrUrVrjryrXrYr{r|s@@rr~r~s# 66,   rr~cxaa]tRt^toRtV3RlV3RlltV3RlRltV3RlRltV3RlR ltR t Vt V;t #) SuperPointInterestPointDecodera The SuperPointInterestPointDecoder uses the output of the SuperPointEncoder to compute the keypoint with scores. The scores are first computed by a convolutional layer, then a softmax is applied to get a probability distribution over the 65 possible keypoint classes. The keypoints are then extracted from the scores by thresholding and non-maximum suppression. Post-processing is then applied to remove keypoints too close to the image borders as well as to keep only the k keypoints with highest score. c$<V^8dQhRS[RR/#rr)rrPs"rr+SuperPointInterestPointDecoder.__annotate__  / D rc<\SV`4VPVnVPVnVPVnVP Vn\ P!RR7Vn\ P!^^R7Vn \ P!VPR,VP^^^R7Vn\ P!VPVP^^^R7VnR#Trfrhr1Nr)rirjkeypoint_threshold max_keypointsr.border_removal_distancerrnrorprqrkrdecoder_hidden_size conv_score_akeypoint_decoder_dim conv_score_brrr^rss&&rrj'SuperPointInterestPointDecoder.__init__s "(";";#11 ++'-'E'E$GGD) LLQq9 II  ' ' +  & &  II  & &(C(CQR[\fg rcx<V^8dQhRS[PRS[S[PS[P3,/#r encodedrrrr)rrPs"rrrs1!!u||!ellELL6P0Q!rcPVPV4pVPV4wr2W23#rx)_get_pixel_scores_extract_keypoints)rrrrr s&& rry&SuperPointInterestPointDecoder.forwards.''0 33F;   rcN<V^8dQhRS[PRS[P/#rrv)rrPs"rrrs#   %,, rc VPVPV44pVPV4p\PP V^4RRR13,pVP wr4rVVP^^^^4PW5V^^4pVP^^^^^4PW5^,V^,4p\W P4pV#)zKBased on the encoder output, compute the scores for each pixel of the imagerNr) rorrrr5softmaxshapepermutereshaperEr.)rrrr batch_sizerArrs&& rr0SuperPointInterestPointDecoder._get_pixel_scoress4,,W56""6*&&vq1!SbS&9'-||$ v1a+33JqRST1aA.66zA:uWXyYFOO4 rcx<V^8dQhRS[PRS[S[PS[P3,/#)r rrr)rrPs"rrrs1!!!% ell@Z:[!rc VPwr#p\P!V^,VP84pV^,\ VP 44,p\ WQVPV^,V^,4wrQVP^8d\WQVP4wrQ\P!V^.4PVP4pWQ3#)z Based on their scores, extract the pixels that represent the keypoints that will be used for descriptors computation. The keypoints are in the form of relative (x, y) coordinates. ) rrnonzerorrtr$rrr,fliptodtype)rrrrArrr s&& rr1SuperPointInterestPointDecoder._extract_keypointss "<<5MM&)d.E.E"EF 5/0: t;;VaZQR     " / 4CUCU V IJJy1#.11&,,?   r)rrrrrr.rqro) rRrSrTrUrVrjryrrrXrYr{r|s@@rrrs7  (!!  !!!rrctaa]tRtRtoRtV3RlV3RlltV3RlRlt]R V3RlRll4tR t Vt V;t #) SuperPointDescriptorDecoderiaS The SuperPointDescriptorDecoder uses the outputs of both the SuperPointEncoder and the SuperPointInterestPointDecoder to compute the descriptors at the keypoints locations. The descriptors are first computed by a convolutional layer, then normalized to have a norm of 1. The descriptors are then interpolated at the keypoints locations. c$<V^8dQhRS[RR/#rr)rrPs"rr(SuperPointDescriptorDecoder.__annotate__rrct<\SV`4\P!RR7Vn\P !^^R7Vn\P!VPR,VP^^^R7Vn \P!VPVP^^^R7Vn R#r) rirjrrnrorprqrkrrconv_descriptor_adescriptor_decoder_dimconv_descriptor_brs&&rrj$SuperPointDescriptorDecoder.__init__s GGD) LLQq9 !#  ' ' +  & & " "$  & &  ) ) " rch<V^8dQhRS[PRS[PRS[P/#)r rr rrv)rrPs"rrr#s.  u||   rc 6VPVPVPV444p\PP V^^R7pVP VR,V^,R,^4^,p\P!V^^4pV#)zXBased on the encoder output and the keypoints, compute the descriptors for each keypointpr(N) rrorrr5 normalize_sample_descriptorsr transpose)rrrr rKs&&& rry#SuperPointDescriptorDecoder.forward#s,,TYYt7M7Mg7V-WX mm--kQA-F ..y At@TVWXYZ[ ook1a8 rc:<V^8dQhRS[RS[P/#)r scaler)rrr)rrPs"rrr0s3u||rc VPwr4rVW^, , R,p\P!Wb,V^, , R, WR,V^, , R, ..4pVPV4pW,pV^,^, pRR/pVP V^R^4p\ P P!W3RR/VBpVPW4R4p\ P PV^^R7pV#)z-Interpolate descriptors at keypoint locationsg? align_cornersTmodebilinearrr) rrtensorrviewrr5 grid_samplerr) r rKrr num_channelsrrdivisorkwargss &&& rr/SuperPointDescriptorDecoder._sample_descriptors/s3>2C2C/ & )C/ ,,%-%!)";c"AV^V[^_V_E_beEe ghi**Y' MA% !4(NN:q"a8 mm// bZb[ab !))*BG mm--kQA-F r)rrrqro)) rRrSrTrUrVrjry staticmethodrrXrYr{r|s@@rrrs<  (  rrcNa]tRtRtoRtRtR tRtV3RlRltV3Rlt Rt Vt R #) SuperPointPreTrainedModeliB superpoint pixel_valuesFcN<V^8dQhRS[PRS[P/#)r rr)rrM)rrPs"rr&SuperPointPreTrainedModel.__annotate__Js(77U=N=N7SXSdSd7rc "VR,R,#)a Assuming pixel_values has shape (batch_size, 3, height, width), and that all channels values are the same, extract the first channel value to get a tensor of shape (batch_size, 1, height, width) for SuperPoint. This is a workaround for the issue discussed in : https://github.com/huggingface/transformers/pull/25786#issuecomment-1730176446 Args: pixel_values: torch.FloatTensor of shape (batch_size, 3, height, width) Returns: pixel_values: torch.FloatTensor of shape (batch_size, 1, height, width) )rrrr)rNrrr )rrrs&&r extract_one_channel_pixel_values:SuperPointPreTrainedModel.extract_one_channel_pixel_valuesJsJ' 66rc&<V^8dQh/S[;R&#)r r^r)rrPs"rrrBs rr N)image) rRrSrTrUbase_model_prefixmain_input_nameinput_modalitiessupports_gradient_checkpointingrrWrXrYrZs@rrrBs/%$O!&+#77rrz@ SuperPoint model outputting keypoints and descriptors. cbaa]tRtRtoRtV3RlV3Rllt]RV3RlRll4tRtVt V;t #) SuperPointForKeypointDetectioni[a SuperPoint model. It consists of a SuperPointEncoder, a SuperPointInterestPointDecoder and a SuperPointDescriptorDecoder. SuperPoint was proposed in `SuperPoint: Self-Supervised Interest Point Detection and Description `__ by Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. It is a fully convolutional neural network that extracts keypoints and descriptors from an image. It is trained in a self-supervised manner, using a combination of a photometric loss and a loss based on the homographic adaptation of keypoints. It is made of a convolutional encoder and two decoders: one for keypoints and one for descriptors. c$<V^8dQhRS[RR/#rr)rrPs"rr+SuperPointForKeypointDetection.__annotate__js  / D rc<\SV`V4Wn\V4Vn\ V4Vn\V4VnVP4R#rx) rirjr^r~encoderrkeypoint_decoderrdescriptor_decoder post_initrs&&rrj'SuperPointForKeypointDetection.__init__jsE   (0 >v F"=f"E rc <V^8dQhRS[PRS[PR,RS[R,RS[R,RS[S[,/#)r rlabelsNrrr)rrM LongTensorrbrrH)rrPs"rrrvsc[ [ ''[   4'[ #Tk [ D[ [  4 4[ rc RpVe \R4hVeTMVPPpVeTMVPPpVP V4pVP wrxrVP VVVR7p V ^,p V U u.uFqPV R ,4NK p p V Uu.uF q^,NK ppV Uu.uF q^,NK pp\X V4U Uu.uF&wp pVPV R ,VR ,4NK( pp p\RV44p\P!VV^3VPR7p\P!VV3VPR7p\P!VVVPP3VPR7p\P!VV3VP\PR7p\!\VVV44FrwpwpppVVVRVP ^,13&VVVRVP ^,13&VVVRVP ^,13&^VVRVP ^,13&Kt V\P"!W.VPR7, pV'd V ^,MRpV'g9\$;QJd.RVVVVVV34FNK 5#!RVVVVVV344#\'VVVVVVR7#uup iuupiuupiuupp i) ao Examples: ```python >>> from transformers import AutoImageProcessor, SuperPointForKeypointDetection >>> import torch >>> from PIL import Image >>> import httpx >>> from io import BytesIO >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> with httpx.stream("GET", url) as response: ... image = Image.open(BytesIO(response.read())) >>> processor = AutoImageProcessor.from_pretrained("magic-leap-community/superpoint") >>> model = SuperPointForKeypointDetection.from_pretrained("magic-leap-community/superpoint") >>> inputs = processor(image, return_tensors="pt") >>> outputs = model(**inputs) ```Nz-SuperPoint does not support training for now.)rrc3F"TFqP^,xK R#5i)rN)r)rr s& rr9SuperPointForKeypointDetection.forward..s#W9OOA$6$6s!)device)rrc30"TF qfKVxK R#5irxr rs& rrrsq$_q$_r)rJr rrKr#rL)N.)r:r^rrrrrrziprmaxrr?rrr enumeraterrrH)rrrrrrrrJrrArrencoder_outputsrlist_keypoints_scoreskeypoints_scoreslist_keypoints list_scoresr list_descriptorsmaximum_num_keypointsrrKr#r _keypoints_scores _descriptorsrLs&&&&&, rry&SuperPointForKeypointDetection.forwardus3:  LM M%9$D $++JjJj &1% ?Yj ! G\\F[2B1--F[\CXYCX/?**CX Y144E~0V 0V,!9  # #$5i$@)IBV W0V  !$#W#W WKK-BA F|ObObc j*?@I\I\]kk . 0R0R S&& {{J(=>|GZGZbgbkbkl6?NT_aq@r6s 2A2 G\2rs ! (U   H %)&0$ :+: :6"))q  q h ,-H Ir