o oiS9@sdZddlZddlmZddlmZmZmZddlZddl m Z m Z m Z m Z ddlmZddlmZmZmZmZmZmZddlmZmZmZd gZGd d d e ZdS) z!Module containing RANSAC modules.N)partial)CallableOptionalTuple)DeviceModuleTensorzeros)KORNIA_CHECK_SHAPE)find_fundamentalfind_homography_dltfind_homography_dlt_iteratedfind_homography_lines_dlt"find_homography_lines_dlt_iteratedsymmetrical_epipolar_distance)#line_segment_transfer_error_one_wayoneway_transfer_errorsample_is_valid_for_homographyRANSACcsbeZdZdZ      d/ded ed ed ed ed eddffdd Zd0deded edee de f ddZ e dededededef ddZ de de de fddZde de d e d edee e eff d!d"Zde de dee e ffd#d$Zd e de fd%d&Zde de d'e de fd(d)Zd0de de d*ee ddfd+d,Zd0de de d*ee dee e ffd-d.ZZS)1raModule for robust geometry estimation with RANSAC. https://en.wikipedia.org/wiki/Random_sample_consensus. Args: model_type: type of model to estimate: "homography", "fundamental", "fundamental_7pt", "homography_from_linesegments". inliers_threshold: threshold for the correspondence to be an inlier. batch_size: number of generated samples at once. max_iterations: maximum batches to generate. Actual number of models to try is ``batch_size * max_iterations``. confidence: desired confidence of the result, used for the early stopping. max_local_iterations: number of local optimization (polishing) iterations. homography@ Gz? model_typeinl_th batch_sizemax_iter confidence max_lo_itersreturnNcstgd|_||_||_||_||_||_||_||_||||dkr7t |_ t |_ t |_d|_d S|dkrIt|_ t|_ t|_d|_d S|dkr[t|_ t|_ d|_t|_d S|dkrqt|_ ttdd |_ d |_t|_d St|d |j) aeInitialize the RANSAC estimator. Args: model_type: type of model to estimate: "homography", "fundamental", "fundamental_7pt", "homography_from_linesegments". inl_th: threshold for the correspondence to be an inlier. batch_size: number of generated samples at once. max_iter: maximum batches to generate. Actual number of models to try is ``batch_size * max_iter``. confidence: desired confidence of the result, used for the early stopping. max_lo_iters: number of local optimization (polishing) iterations. )r fundamentalfundamental_7pthomography_from_linesegmentsrr$r"r#7POINT)methodz is unknown. Try one of N)super__init__supported_modelsrrrrrr rerror_fnr minimal_solverr polisher_solverminimal_sample_sizerrrrr rNotImplementedError)selfrrrrrr  __class__J/home/ubuntu/.local/lib/python3.10/site-packages/kornia/geometry/ransac.pyr+;sB      zRANSAC.__init__ sample_sizepop_sizedevicecCs8|dur td}tj|||d}|j|dd\}}|S)aMinimal sampler, but unlike traditional RANSAC we sample in batches. Yields the benefit of the parallel processing, esp. on GPU. Args: sample_size: number of samples to draw from the population. pop_size: size of the population to sample from. batch_size: number of sample sets to generate. device: device to place the samples on. Returns: Tensor of sampled indices with shape :math:`(batch_size, sample_size)`. Ncpu)r9)kdim)torchr9randtopk)r2r7r8rr9r?_outr5r5r6sampleus  z RANSAC.samplen_inlnum_tcconfc CsRd}||krdS||krdStd|t| tt|dt|||S)aUpdate max_iter to stop iterations earlier https://en.wikipedia.org/wiki/Random_sample_consensus. Args: n_inl: number of inliers. num_tc: total number of correspondences. sample_size: size of minimal sample. conf: desired confidence level. Returns: Maximum number of samples needed to achieve the desired confidence. g& .>g?)mathlogminmaxpow)rDrEr7rFepsr5r5r6max_samples_by_confs 6zRANSAC.max_samples_by_confkp1kp2c Cs6|jdd\}}|||tj|||j|jd}|S)aEstimate models from minimal samples. Args: kp1: source keypoints with shape :math:`(batch_size, sample_size, 2)`. kp2: target keypoints with shape :math:`(batch_size, sample_size, 2)`. Returns: Estimated models tensor. Ndtyper9)shaper.r>onesrRr9)r2rNrOrr7Hr5r5r6estimate_model_from_minsamples z$RANSAC.estimate_model_from_minsamplemodelsc Cst|jdkr |d}t|jdkr|d}|jd}|jdkr4|||ddd||ddd|}n|||dd||dd|}||k}||jdd}|} || } ||  } || } | | | fS)aVerify models by computing inliers and selecting the best model. Args: kp1: source keypoints. kp2: target keypoints. models: candidate models to verify. inl_th: inlier threshold. Returns: Tuple containing: - Best model - Inlier mask for the best model - Score of the best model rPNrr$r;r=) lenrSrr-expandtosumargmaxitemclone) r2rNrOrWrrerrorsinl models_scorebest_model_idxbest_model_score model_best inliers_bestr5r5r6verifys  ("   z RANSAC.verifycCs,|jdkrt||}||||fS||fS)zRemove degenerate samples based on model-specific constraints. Args: kp1: source keypoints. kp2: target keypoints. Returns: Tuple of filtered keypoints (kp1, kp2). r)rr)r2rNrOmaskr5r5r6remove_bad_sampless zRANSAC.remove_bad_samplescCs0tj|ddd}|jddddk}||S)zRemove degenerate models based on simple heuristics. Args: models: candidate models to filter. Returns: Filtered models tensor. r;rP)dim1dim2rYrg-C6?)r>diagonalabsrI)r2rW main_diagonalrir5r5r6remove_bad_modelss zRANSAC.remove_bad_modelsinliersc CsF||d}||d}|d}|||tjd||j|jd}|S)aPolish the model using inliers through local optimization. Args: kp1: source keypoints. kp2: target keypoints. inliers: boolean mask indicating inlier correspondences. Returns: Polished model tensor. Nr;rQ)sizer/r>rTrRr9)r2rNrOrqkp1_inlkp2_inlnum_inlmodelr5r5r6 polish_models  zRANSAC.polish_modelweightscCs|jdvr5t|ddgt|ddg|jd|jdkr%|jd|jkr5td|jd|jd|j|jdkrjt|gd t|gd |jd|jdkrZ|jd|jkrltd|jd |jd|jd Sd S) aGValidate input tensors for shape and size requirements. Args: kp1: source keypoints. kp2: target keypoints. weights: optional correspondence weights (not used currently). Raises: ValueError: if input shapes are invalid or insufficient correspondences. )rr"N2rzMkp1 and kp2 should be equal shape at least [z+, 2], got z, r$)ryrzrzz., 2, 2], got N)rr rSr0 ValueError)r2rNrOrxr5r5r6validate_inputss8 $ $zRANSAC.validate_inputscCs||||t|j}t|}tdd|j|jd}t|d|jtjd}t |j D]}| |j||j |j} || } || } | | | \} } t| dkrMq)|| | } || } | dusbt| dkrcq)|||| |j\} }}||krt |jD]1}||||}|dust|dkrqx|||||j\}}}||kr|d} |}|}qx| }|}|}t|t|||j|j}|d|j |kr||fSq)||fS)aCall main forward method to execute the RANSAC algorithm. Args: kp1: source image keypoints :math:`(N, 2)`. kp2: distance image keypoints :math:`(N, 2)`. weights: optional correspondences weights. Not used now. Returns: - Estimated model, shape of :math:`(1, 3, 3)`. - The inlier/outlier mask, shape of :math:`(1, N)`, where N is number of input correspondences. rQr;)r9rRrN)r|floatr0rZr rRr9r>boolrangerrCrrjrVrprhrr rwr`intrMr)r2rNrOrxbest_score_totalrEbest_model_totalinliers_best_totaliidxs kp1_sampled kp2_sampledrWrvrq model_scorerAmodel_lo inliers_loscore_lo new_max_iterr5r5r6forward$sN     zRANSAC.forward)rrrrrr)N)__name__ __module__ __qualname____doc__strr~rr+rrrrC staticmethodrMrVrrhrjrprwr|r __classcell__r5r5r3r6r-sB$: (! 0)rrG functoolsrtypingrrrr> kornia.corerrrr kornia.core.checkr kornia.geometryr r r rrrkornia.geometry.homographyrrr__all__rr5r5r5r6s