o #i@sddlZddlZddlZddlZddlZddlmZmZddlmZddl m Z ddl m Z m Z mZmZddlZddlmZddlmZmZmZdd lmZeejejeejejfZeejejeejfZd Zejedd Z Gd d d eeZ!Gddde!Z"Gddde!Z#Gddde!Z$Gddde!Z%Gddde!Z&Gddde!Z'Gddde!Z(Gddde!Z)Gddde!Z*Gd d!d!e!Z+dS)"N)ABCabstractmethod)glob)Path)CallablecastOptionalUnion)Image) _read_pfmdownload_and_extract_archiveverify_str_arg) VisionDataset)slice_channelsc seZdZdZdZddeeefdee ddffdd Z d eeefde j fd d Z dd ed eede eeeeffddZed edeeejeejffddZdedeeeffddZdefddZZS)StereoMatchingDatasetz+Base interface for Stereo matching datasetsFNroot transformsreturncs$tj|d||_g|_g|_dS)a} Args: root(str): Root directory of the dataset. transforms(callable, optional): A function/transform that takes in Tuples of (images, disparities, valid_masks) and returns a transformed version of each of them. images is a Tuple of (``PIL.Image``, ``PIL.Image``) disparities is a Tuple of (``np.ndarray``, ``np.ndarray``) with shape (1, H, W) valid_masks is a Tuple of (``np.ndarray``, ``np.ndarray``) with shape (H, W) In some cases, when a dataset does not provide disparities, the ``disparities`` and ``valid_masks`` can be Tuples containing None values. For training splits generally the datasets provide a minimal guarantee of images: (``PIL.Image``, ``PIL.Image``) disparities: (``np.ndarray``, ``None``) with shape (1, H, W) Optionally, based on the dataset, it can return a ``mask`` as well: valid_masks: (``np.ndarray | None``, ``None``) with shape (H, W) For some test splits, the datasets provides outputs that look like: imgaes: (``PIL.Image``, ``PIL.Image``) disparities: (``None``, ``None``) Optionally, based on the dataset, it can return a ``mask`` as well: valid_masks: (``None``, ``None``) rN)super__init__r_images _disparities)selfrr __class__rb/home/ubuntu/veenaModal/venv/lib/python3.10/site-packages/torchvision/datasets/_stereo_matching.pyrs zStereoMatchingDataset.__init__ file_pathcCs"t|}|jdkr|d}|S)NRGB)r openmodeconvert)rrimgrrr _read_img:s   zStereoMatchingDataset._read_imgpaths_left_patternpaths_right_patternc Csttt|}|rttt|}n tdd|D}|s%td||s.td|t|t|krKtdt|dt|d|d|d td dt||D}|S) Ncss|]}dVqdSNr.0_rrr Lz4StereoMatchingDataset._scan_pairs..z0Could not find any files matching the patterns: zFound z left files but z# right files using: left pattern: z right pattern:  css|] \}}||fVqdSr(r)r*leftrightrrrr,[s)listsortedrFileNotFoundErrorlen ValueErrorzip)rr&r' left_paths right_pathspathsrrr _scan_pairs@s$z!StereoMatchingDataset._scan_pairscCsdSr(r)rrrrr_read_disparity^sz%StereoMatchingDataset._read_disparityindexc Cs||j|d}||j|d}||j|d\}}||j|d\}}||f}||f} ||f} |jdurG||| | \}} } |jsP| ddurb|d|d| dttj| dfS|d|d| dfS)aoReturn example at given index. Args: index(int): The index of the example to retrieve Returns: tuple: A 3 or 4-tuple with ``(img_left, img_right, disparity, Optional[valid_mask])`` where ``valid_mask`` can be a numpy boolean mask of shape (H, W) if the dataset provides a file indicating which disparity pixels are valid. The disparity is a numpy array of shape (1, H, W) and the images are PIL images. ``disparity`` is None for datasets on which for ``split="test"`` the authors did not provide annotations. rr N) r%rr;rr_has_built_in_disparity_maskrnpndarray) rr<img_left img_right dsp_map_leftvalid_mask_left dsp_map_rightvalid_mask_rightimgsdsp_maps valid_masksrrr __getitem__cs    $z!StereoMatchingDataset.__getitem__cCs t|jSr()r4r)rrrr__len__s zStereoMatchingDataset.__len__r()__name__ __module__ __qualname____doc__r=r strrrrrr r%r1tupler:rr>r?r;intT1T2rIrJ __classcell__rrrrrs"(  (#rcsneZdZdZddeeefdeeddffdd Z dede e j dffd d Z d edeffd d ZZS) CarlaStereoaz Carla simulator data linked in the `CREStereo github repo `_. The dataset is expected to have the following structure: :: root carla-highres trainingF scene1 img0.png img1.png disp0GT.pfm disp1GT.pfm calib.txt scene2 img0.png img1.png disp0GT.pfm disp1GT.pfm calib.txt ... Args: root (str or ``pathlib.Path``): Root directory where `carla-highres` is located. transforms (callable, optional): A function/transform that takes in a sample and returns a transformed version. Nrrrc st||t|d}t|ddd}t|ddd}|||}||_t|ddd}t|ddd}|||}||_dS)Nz carla-highres trainingF*im0.pngim1.png disp0GT.pfmz disp1GT.pfmrrrrOr:rr) rrrleft_image_patternright_image_patternrFleft_disparity_patternright_disparity_pattern disparitiesrrrrs    zCarlaStereo.__init__rcCt|}t|}d}||fSr(_read_pfm_filer>absrr disparity_map valid_maskrrrr; zCarlaStereo._read_disparityr<cttt|SaReturn example at given index. Args: index(int): The index of the example to retrieve Returns: tuple: A 3-tuple with ``(img_left, img_right, disparity)``. The disparity is a numpy array of shape (1, H, W) and the images are PIL images. If a ``valid_mask`` is generated within the ``transforms`` parameter, a 4-tuple with ``(img_left, img_right, disparity, valid_mask)`` is returned. rrRrrIrr<rrrrI zCarlaStereo.__getitem__r(rKrLrMrNr rOrrrrrPr>r?r;rQrRrIrTrrrrrUs (rUc zeZdZdZdZddeeefdedee ddffd d Z d ede ee j dffd d Zdedeffdd ZZS)Kitti2012Stereoa  KITTI dataset from the `2012 stereo evaluation benchmark `_. Uses the RGB images for consistency with KITTI 2015. The dataset is expected to have the following structure: :: root Kitti2012 testing colored_0 1_10.png 2_10.png ... colored_1 1_10.png 2_10.png ... training colored_0 1_10.png 2_10.png ... colored_1 1_10.png 2_10.png ... disp_noc 1.png 2.png ... calib Args: root (str or ``pathlib.Path``): Root directory where `Kitti2012` is located. split (string, optional): The dataset split of scenes, either "train" (default) or "test". transforms (callable, optional): A function/transform that takes in a sample and returns a transformed version. TtrainNrsplitrrcst||t|dddt|d|d}t|dd}t|dd}||||_|d krDt|d d }||d|_dStd d |jD|_dS)Nrrrqtest valid_values Kitti2012ing colored_0z*_10.png colored_1rqdisp_noc*.pngcs|]}dVqdSNNNrr)rrrr,r-z+Kitti2012Stereo.__init__.. rrrrrOr:rrr1)rrrrrleft_img_patternright_img_patterndisparity_patternrrrrszKitti2012Stereo.__init__rcCB|durdStt|d}|dddddf}d}||fSNrgp@r>asarrayr r!rerrrr; zKitti2012Stereo._read_disparityr<criaReturn example at given index. Args: index(int): The index of the example to retrieve Returns: tuple: A 4-tuple with ``(img_left, img_right, disparity, valid_mask)``. The disparity is a numpy array of shape (1, H, W) and the images are PIL images. ``valid_mask`` is implicitly ``None`` if the ``transforms`` parameter does not generate a valid mask. Both ``disparity`` and ``valid_mask`` are ``None`` if the dataset split is test. rkrlrrrrI zKitti2012Stereo.__getitem__rqNrKrLrMrNr=r rOrrrrrPr>r?r;rQrRrIrTrrrrrps &,  rpc ro)Kitti2015StereoaM KITTI dataset from the `2015 stereo evaluation benchmark `_. The dataset is expected to have the following structure: :: root Kitti2015 testing image_2 img1.png img2.png ... image_3 img1.png img2.png ... training image_2 img1.png img2.png ... image_3 img1.png img2.png ... disp_occ_0 img1.png img2.png ... disp_occ_1 img1.png img2.png ... calib Args: root (str or ``pathlib.Path``): Root directory where `Kitti2015` is located. split (string, optional): The dataset split of scenes, either "train" (default) or "test". transforms (callable, optional): A function/transform that takes in a sample and returns a transformed version. TrqNrrrrrcst||t|dddt|d|d}t|dd}t|dd}||||_|d krLt|d d}t|d d}||||_dStd d |jD|_dS)Nrrrsru Kitti2015rximage_2r|image_3rq disp_occ_0 disp_occ_1csr}r~rr)rrrr,Zr-z+Kitti2015Stereo.__init__..rrrrrrrrr^r_rrrrKszKitti2015Stereo.__init__rcCrrrrerrrr;\rzKitti2015Stereo._read_disparityr<crirrkrlrrrrIgrzKitti2015Stereo.__getitem__rrrrrrrs ),  rcseZdZdZgdgdgddZdZ   dd eeefd ed e ede de e de dd ffdd Z deeefde j ffdd ZdedeedeejejfffddZd eeefdd fddZdedeffdd ZZS) Middlebury2014StereoaZ Publicly available scenes from the Middlebury dataset `2014 version `. The dataset mostly follows the original format, without containing the ambient subdirectories. : :: root Middlebury2014 train scene1-{perfect,imperfect} calib.txt im{0,1}.png im1E.png im1L.png disp{0,1}.pfm disp{0,1}-n.png disp{0,1}-sd.pfm disp{0,1}y.pfm scene2-{perfect,imperfect} calib.txt im{0,1}.png im1E.png im1L.png disp{0,1}.pfm disp{0,1}-n.png disp{0,1}-sd.pfm disp{0,1}y.pfm ... additional scene1-{perfect,imperfect} calib.txt im{0,1}.png im1E.png im1L.png disp{0,1}.pfm disp{0,1}-n.png disp{0,1}-sd.pfm disp{0,1}y.pfm ... test scene1 calib.txt im{0,1}.png scene2 calib.txt im{0,1}.png ... Args: root (str or ``pathlib.Path``): Root directory of the Middleburry 2014 Dataset. split (string, optional): The dataset split of scenes, either "train" (default), "test", or "additional" use_ambient_views (boolean, optional): Whether to use different expose or lightning views when possible. The dataset samples with equal probability between ``[im1.png, im1E.png, im1L.png]``. calibration (string, optional): Whether or not to use the calibrated (default) or uncalibrated scenes. transforms (callable, optional): A function/transform that takes in a sample and returns a transformed version. download (boolean, optional): Whether or not to download the dataset in the ``root`` directory. ) Adirondack Jadeplant MotorcyclePianoPipesPlayroom PlaytableRecycleShelvesVintage) BackpackBicycle1Cable Classroom1CouchFlowersMaskShopvacSticksStorageSword1Sword2Umbrella)Plants Classroom2E Classroom2 AustraliaDjembeLCrusadePCrusadeHoopsBicycle2 StaircaseNewkuba AustraliaPDjembe LivingroomComputer)rq additionalrtTrqperfectFNrrr calibrationuse_ambient_viewsrdownloadrcst||t|ddd||_|r#t|ddd|dkr"tdn|dkr2td|d |d |r9||t|d }tj ||sOt d |d |j |t fddt ||Dslt d|ddgdgdgddgd|}|D]T}d|} t||| d} t||| d} |j|| | 7_|dkrtdd|jD|_q|t||| d} t||| d} |j|| | 7_q|||_dS)Nrr)rqrtrrur)r imperfectbothNrtzMSplit 'test' has only no calibration settings, please set `calibration=None`.zSplit 'zr' has calibration settings, however None was provided as an argument. Setting calibration to 'perfect' for split 'zF'. Available calibration settings are: 'perfect', 'imperfect', 'both'.Middlebury2014zThe z7 directory was not found in the provided root directoryc3s$|] }D]}||VqqdSr() startswith)r*scenes split_scenesrrr,s z0Middlebury2014Stereo.__init__..z:Provided root folder does not contain any scenes from the z split.z-perfectz -imperfect)NrrrrWrXrYcsr}r~rr)rrrr,r-z disp0.pfmz disp1.pfm)rrrrrr5_download_datasetrospathexistsr3splitsanylistdirrOrr:r1rr)rrrrrrrrcalibrartion_suffixescalibration_suffix scene_patternrrleft_dispartity_patternright_dispartity_patternrrrrsT       zMiddlebury2014Stereo.__init__rcspt|ts t|}|jdkr2|jr2|jtfdddD}ttdd|}||t |}t |S)a Function that reads either the original right image or an augmented view when ``use_ambient_views`` is True. When ``use_ambient_views`` is True, the dataset will return at random one of ``[im1.png, im1E.png, im1L.png]`` as the right image. rYc3s|]}|VqdSr(r)r* view_name base_pathrrr,0sz1Middlebury2014Stereo._read_img..)zim1E.pngzim1L.pngcSs tj|Sr()rrr)prrr2s z0Middlebury2014Stereo._read_img..) isinstancernamerparentr1filterappendrandomchoicerr%)rrambient_file_pathsrrrr%"s    zMiddlebury2014Stereo._read_imgrcCsB|durdSt|}t|}d||tjk<|dkd}||fS)Nrr)rcr>rdinfsqueezererrrr;8s z$Middlebury2014Stereo._read_disparitycs@d}td|j}|dkrD|j|D]-}|}dD]$}|d|}|d|d}||s@t||dt|dd qqdStdtfd d |jdDrd } t| tdd t tdD]%\} } } | D]} d}t| | }tj|ddt t|t|qtqmt tddSdS)Nz8https://vision.middlebury.edu/stereo/data/scenes2014/ziprrt)rr-/z.zipT)urlfilename download_rootremove_finishedc3s"|] }|tdvVqdS)rtN)rr)r*rrrrr,Ys z9Middlebury2014Stereo._download_dataset..zEhttps://vision.middlebury.edu/stereo/submit3/zip/MiddEval3-data-F.zip)rrrzMiddEval3/testF)exist_ok MiddEval3) rrrrrr rOrmakedirsrwalkshutilmovermtree)rrbase_url split_name split_scene split_rootr scene_name scene_url test_set_url scene_dir scene_namesr+r scene_dst_dir scene_src_dirrrrrCs@   z&Middlebury2014Stereo._download_datasetr<cri)azReturn example at given index. Args: index(int): The index of the example to retrieve Returns: tuple: A 4-tuple with ``(img_left, img_right, disparity, valid_mask)``. The disparity is a numpy array of shape (1, H, W) and the images are PIL images. ``valid_mask`` is implicitly ``None`` for `split=test`. rrSrrIrlrrrrIi z Middlebury2014Stereo.__getitem__)rqrFNF)rKrLrMrNrr=r rOrrboolrrr r%rPr>r?r;rrQrSrIrTrrrrrws>9 /  A* &rcsteZdZdZdZ ddeeefdee ddffdd Z d ede e j dffd d Zd edeffd d ZZS) CREStereoaSynthetic dataset used in training the `CREStereo `_ architecture. Dataset details on the official paper `repo `_. The dataset is expected to have the following structure: :: root CREStereo tree img1_left.jpg img1_right.jpg img1_left.disp.jpg img1_right.disp.jpg img2_left.jpg img2_right.jpg img2_left.disp.jpg img2_right.disp.jpg ... shapenet img1_left.jpg img1_right.jpg img1_left.disp.jpg img1_right.disp.jpg ... reflective img1_left.jpg img1_right.jpg img1_left.disp.jpg img1_right.disp.jpg ... hole img1_left.jpg img1_right.jpg img1_left.disp.jpg img1_right.disp.jpg ... Args: root (str): Root directory of the dataset. transforms (callable, optional): A function/transform that takes in a sample and returns a transformed version. TNrrrc st||t|d}gd}|D]<}t||d}t||d}|||}|j|7_t||d}t||d} ||| } |j| 7_qdS)Nr)shapenet reflectivetreeholez *_left.jpgz *_right.jpgz*_left.disp.pngz*_right.disp.pngr[) rrrdirsrr\r]rFr^r_r`rrrrs   zCREStereo.__init__rcC<tjt|tjd}|dddddfd}d}||fS)Ndtypeg@@r>rr r!float32rerrrr;zCREStereo._read_disparityr<cri)aReturn example at given index. Args: index(int): The index of the example to retrieve Returns: tuple: A 4-tuple with ``(img_left, img_right, disparity, valid_mask)``. The disparity is a numpy array of shape (1, H, W) and the images are PIL images. ``valid_mask`` is implicitly ``None`` if the ``transforms`` parameter does not generate a valid mask. rkrlrrrrIrmzCREStereo.__getitem__r(rrrrrrws) rc reZdZdZddeeefdedeeddffdd Z d ede e j dffd d Z d edeffdd ZZS)FallingThingsStereoa`FallingThings `_ dataset. The dataset is expected to have the following structure: :: root FallingThings single dir1 scene1 _object_settings.json _camera_settings.json image1.left.depth.png image1.right.depth.png image1.left.jpg image1.right.jpg image2.left.depth.png image2.right.depth.png image2.left.jpg image2.right ... scene2 ... mixed scene1 _object_settings.json _camera_settings.json image1.left.depth.png image1.right.depth.png image1.left.jpg image1.right.jpg image2.left.depth.png image2.right.depth.png image2.left.jpg image2.right ... scene2 ... Args: root (str or ``pathlib.Path``): Root directory where FallingThings is located. variant (string): Which variant to use. Either "single", "mixed", or "both". transforms (callable, optional): A function/transform that takes in a sample and returns a transformed version. singleNrvariantrrc st||t|d}t|ddddgdgddgd|}tddtdd}|D]H}t||||d }t||||d }|j|||7_t||||d } t||||d } |j|| | 7_q-dS) N FallingThingsr)r mixedrrur rrW)r rz *.left.jpgz *.right.jpgz*.left.depth.pngz*.right.depth.pngrrrrrOrr:r) rrrrvariants split_prefixrrrr^r_rrrrs(  zFallingThingsStereo.__init__rc Cstt|}t|jd}t|8}t|}|dddd}d\}}||||tj } | dddddf} d} | | fWdS1sNwYdS)Nz_camera_settings.jsoncamera_settingsrintrinsic_settingsfx)d) r>rr r!rrjsonloadastyper ) rrdepthcamera_settings_pathf intrinsicsfocalbaselinepixel_constantrfrgrrrr;s  $z#FallingThingsStereo._read_disparityr<crirjrkrlrrrrI(rmzFallingThingsStereo.__getitem__)r Nrnrrrrr s ,,r c s|eZdZdZ   ddeeefdededeed df fd d Z d ed e e j dffd dZ ded effdd ZZS)SceneFlowStereoaDataset interface for `Scene Flow `_ datasets. This interface provides access to the `FlyingThings3D, `Monkaa` and `Driving` datasets. The dataset is expected to have the following structure: :: root SceneFlow Monkaa frames_cleanpass scene1 left img1.png img2.png right img1.png img2.png scene2 left img1.png img2.png right img1.png img2.png frames_finalpass scene1 left img1.png img2.png right img1.png img2.png ... ... disparity scene1 left img1.pfm img2.pfm right img1.pfm img2.pfm FlyingThings3D ... ... Args: root (str or ``pathlib.Path``): Root directory where SceneFlow is located. variant (string): Which dataset variant to user, "FlyingThings3D" (default), "Monkaa" or "Driving". pass_name (string): Which pass to use, "clean" (default), "final" or "both". transforms (callable, optional): A function/transform that takes in a sample and returns a transformed version. FlyingThings3DcleanNrr pass_namerrc s(t||t|d}t|dddt|ddddgdgddgd|}||}td td d d td d d d }|D]P}t||||d d }t||||d d } |j||| 7_t|d||d d} t|d||d d} |j|| | 7_qAdS)N SceneFlowr)r$DrivingMonkaarur&)r%finalrframes_cleanpassframes_finalpassrW)r)r$r(r/r|r0 disparityz*.pfmr) rrrr&rpassesprefix_directoriesrr\r]r^r_rrrrms. zSceneFlowStereo.__init__rcCrar(rbrerrrr;rhzSceneFlowStereo._read_disparityr<crirjrkrlrrrrIrmzSceneFlowStereo.__getitem__)r$r%Nrnrrrrr#7s$8 %r#c seZdZdZdZddeeefdedee ddffd d Z d ede eeffd d Z d edee de e je jfffddZdedeffdd ZZS) SintelStereoaSintel `Stereo Dataset `_. The dataset is expected to have the following structure: :: root Sintel training final_left scene1 img1.png img2.png ... ... final_right scene2 img1.png img2.png ... ... disparities scene1 img1.png img2.png ... ... occlusions scene1 img1.png img2.png ... ... outofframe scene1 img1.png img2.png ... ... Args: root (str or ``pathlib.Path``): Root directory where Sintel Stereo is located. pass_name (string): The name of the pass to use, either "final", "clean" or "both". transforms (callable, optional): A function/transform that takes in a sample and returns a transformed version. Tr*Nrr&rrc st||t|dddt|d}dgdgddgd|}|D]B}t|d|dd d }t|d|d d d }|j|||7_t|dd d d }|j||d7_q"dS) Nr&)r*r%rruSintelr*r%training_leftrWr|_rightr`)rrrrrOrr:r) rrr&r pass_namesrrrrrrrrs  zSintelStereo.__init__rcCst|}|j}|j}|jj}t|d|j|}t|d|j|}tj|s2td|dtj|s@td|d||fS)N occlusions outofframezOcclusion mask z does not existzOut of frame mask )rrrrOrrrr3)rrfpathbasenamescenedir sampledirocclusion_pathoutofframe_pathrrr_get_occlussion_mask_pathss  z'SintelStereo._get_occlussion_mask_pathsrc Cs|durdStjt|tjd}tj|ddd\}}}|d|d|d}t|d }||\}}tt|d k}tt|d k} t| |}||fS) Nrr)axis@i@)rr r) r>rr r!r rr transposer> logical_and) rrrfrgbocclued_mask_pathout_of_frame_mask_pathrgoff_maskrrrr;s  zSintelStereo._read_disparityr<cri)aReturn example at given index. Args: index(int): The index of the example to retrieve Returns: tuple: A 4-tuple with ``(img_left, img_right, disparity, valid_mask)`` is returned. The disparity is a numpy array of shape (1, H, W) and the images are PIL images whilst the valid_mask is a numpy array of shape (H, W). rrlrrrrIrzSintelStereo.__getitem__)r*N)rKrLrMrNr=r rOrrrrrPr>r>r?r;rQrSrIrTrrrrr0s,,*r0c r ) InStereo2ka`InStereo2k `_ dataset. The dataset is expected to have the following structure: :: root InStereo2k train scene1 left.png right.png left_disp.png right_disp.png ... scene2 ... test scene1 left.png right.png left_disp.png right_disp.png ... scene2 ... Args: root (str or ``pathlib.Path``): Root directory where InStereo2k is located. split (string): Either "train" or "test". transforms (callable, optional): A function/transform that takes in a sample and returns a transformed version. rqNrrrrrcst||t|d|}t|dddt|dd}t|dd}||||_t|dd}t|dd }||||_dS) NrMrrrsrurWzleft.pngz right.pngz left_disp.pngzright_disp.png)rrrrrOr:rrrrrrrAszInStereo2k.__init__rcCr)Nrg@rrerrrr;Pr zInStereo2k._read_disparityr<crirjrkrlrrrrIWrmzInStereo2k.__getitem__rrnrrrrrM!s ,rMc seZdZdZdZddeeefdedee ddffd d Z d edee d e e j e j fffd dZdedeffdd ZZS) ETH3DStereoafETH3D `Low-Res Two-View `_ dataset. The dataset is expected to have the following structure: :: root ETH3D two_view_training scene1 im1.png im0.png images.txt cameras.txt calib.txt scene2 im1.png im0.png images.txt cameras.txt calib.txt ... two_view_training_gt scene1 disp0GT.pfm mask0nocc.png scene2 disp0GT.pfm mask0nocc.png ... two_view_testing scene1 im1.png im0.png images.txt cameras.txt calib.txt scene2 im1.png im0.png images.txt cameras.txt calib.txt ... Args: root (str or ``pathlib.Path``): Root directory of the ETH3D Dataset. split (string, optional): The dataset split of scenes, either "train" (default) or "test". transforms (callable, optional): A function/transform that takes in a sample and returns a transformed version. TrqNrrrrrc st||t|dddt|d}|dkrdnd}d}t||d d }t||d d }||||_|d krJtd d|jD|_dSt||d d}||d|_dS)NrrrsruETH3Drqtwo_view_training two_view_testtwo_view_training_gtrWrXrYrtcsr}r~rr)rrrr,r-z'ETH3DStereo.__init__..rZ) rrrrrOr:rr1r) rrrrrimg_diranot_dirrrrrrrrs zETH3DStereo.__init__rrcCsN|durdSt|}t|}t|jd}t|}t|t }||fS)Nrz mask0nocc.png) rcr>rdrrr r!rrr)rrrf mask_pathrgrrrr;s  zETH3DStereo._read_disparityr<crirrrlrrrrIrzETH3DStereo.__getitem__r)rKrLrMrNr=r rOrrrrrPr>r?r;rQrSrIrTrrrrrNfs 1,* rN), functoolsrrrrabcrrrpathlibrtypingrrrr numpyr>PILr utilsr r rvisionrrPr?rRrS__all__partialrcrrUrprrrr r#r0rMrNrrrrs<    q@UXXhpzE