U eք@sbdZddlZddlZddlZddlZddlmZmZm Z ddl TGdddZ Gdd d e Z dS) z The source code is adapted from https://github.com/aliaksandr960/segment-anything-eo. Credit to the author Aliaksandr Hancharenka. N)sam_model_registrySamAutomaticMaskGenerator SamPredictor)*c @seZdZdZd;ddZdddZd?ddZd@ddZ dAddZ dde j ddfddZ dBddZdddd ifd!d"ZdCd$d%ZdDd)d*Zd+d,Zd-d.Zd/d0ZdEd1d2ZdFd3d4ZdGd5d6ZdHd7d8ZdId9d:ZdS)JSamGeozThe main class for segmenting geospatial data with the Segment Anything Model (SAM). See https://github.com/facebookresearch/segment-anything for details. vit_hTNc Ksd}d|kr8|d}tj|s,t|||}|dn t|||}|dkrptjrZdnd}|dkrptj||_ ||_ ||_ ||_ d|_ d|_d|_d|_d|_d|_d|_d|_t|j |j d|_|jj|j d|j dk r|j ni}|rt|jf||_nt|jf||_dS)aInitialize the class. Args: model_type (str, optional): The model type. It can be one of the following: vit_h, vit_l, vit_b. Defaults to 'vit_h'. See https://bit.ly/3VrpxUh for more details. automatic (bool, optional): Whether to use the automatic mask generator or input prompts. Defaults to True. The automatic mask generator will segment the entire image, while the input prompts will segment selected objects. device (str, optional): The device to use. It can be one of the following: cpu, cuda. Defaults to None, which will use cuda if available. checkpoint_dir (str, optional): The path to the model checkpoint. It can be one of the following: sam_vit_h_4b8939.pth, sam_vit_l_0b3195.pth, sam_vit_b_01ec64.pth. Defaults to None. See https://bit.ly/3VrpxUh for more details. sam_kwargs (dict, optional): Optional arguments for fine-tuning the SAM model. Defaults to None. The available arguments with default values are listed below. See https://bit.ly/410RV0v for more details. points_per_side: Optional[int] = 32, points_per_batch: int = 64, pred_iou_thresh: float = 0.88, stability_score_thresh: float = 0.95, stability_score_offset: float = 1.0, box_nms_thresh: float = 0.7, crop_n_layers: int = 0, crop_nms_thresh: float = 0.7, crop_overlap_ratio: float = 512 / 1500, crop_n_points_downscale_factor: int = 1, point_grids: Optional[List[np.ndarray]] = None, min_mask_region_area: int = 0, output_mode: str = "binary_mask", F checkpointNcudacpu)r device)ospathexistsZdownload_checkpointpoptorchr is_available empty_cacher model_typer sam_kwargssourceimagemasksobjects annotations predictionscoreslogitsrZsamtormask_generatorr predictor) selfrZ automaticr Zcheckpoint_dirrkwargsZhqr r$P/home/ankuromar296_gmail_com/.local/lib/python3.8/site-packages/samgeo/samgeo.py__init__s:'     zSamGeo.__init__r(cKs|j\}}}|j|} |r2tj||ftjd} ntj||ftjd} tj||ftjd} | D]X} | ddktj} | | 7} |dk r^tj | |dd}|dktj}| |}| |7} q^| dktj} | dktj} | | }||S)aGenerate masks for the input tile. This function originates from the segment-anything-eo repository. See https://bit.ly/41pwiHw Args: image (np.ndarray): The input image as a numpy array. foreground (bool, optional): Whether to generate the foreground mask. Defaults to True. erosion_kernel (tuple, optional): The erosion kernel for filtering object masks and extract borders. Defaults to (3, 3). mask_multiplier (int, optional): The mask multiplier for the output mask, which is usually a binary mask [0, 1]. You can use this parameter to scale the mask to a larger range, for example [0, 255]. Defaults to 255. dtype segmentationrNrZ iterations) shaper generatenpzerosuint8onesastypecv2erode)r"r foregrounderosion_kernelmask_multiplierr#hw_rresulting_maskresulting_bordersmmask mask_erode edge_maskZresulting_mask_with_bordersr$r$r%__call__is$   zSamGeo.__call__Fc Kst|tr|drt|}tj|s8td|d|rjd|_||_ ||_ t |||f|||d|St |} t | t j} nt|tjr|} d}ntd||_ | |_|j} | | } | |_ d|_|dk r|j|||||f|dS) aGenerate masks for the input image. Args: source (str | np.ndarray): The path to the input image or the input image as a numpy array. output (str, optional): The path to the output image. Defaults to None. foreground (bool, optional): Whether to generate the foreground mask. Defaults to True. batch (bool, optional): Whether to generate masks for a batch of image tiles. Defaults to False. erosion_kernel (tuple, optional): The erosion kernel for filtering object masks and extract borders. Such as (3, 3) or (5, 5). Set to None to disable it. Defaults to None. mask_multiplier (int, optional): The mask multiplier for the output mask, which is usually a binary mask [0, 1]. You can use this parameter to scale the mask to a larger range, for example [0, 255]. Defaults to 255. The parameter is ignored if unique is True. unique (bool, optional): Whether to assign a unique value to each object. Defaults to True. The unique value increases from 1 to the number of objects. The larger the number, the larger the object area. http Input path  does not exist.T)r7r8r9Nz4Input source must be either a path or a numpy array.F) isinstancestr startswith download_filerrr ValueErrorbatchrrZ tiff_to_tiffr5imreadcvtColor COLOR_BGR2RGBr0ndarrayrr r/ save_masks) r"routputr7rLr8r9uniquer#rr rr$r$r%r/sT      zSamGeo.generatecKs|jdkrtd|jj\}}} |j} t| dkr:tj} nt| dkrNtj} ntj} |rt | dddd} t | d d jd | d d jd f} t | D]\}}|d }|d | |<qn|rtj ||f| d }ntj ||f| d }tj ||f| d }| D]T}|d d k | }||7}|dk rtj||d d }|d k | }||}||7}q|d k | }|d k | }||} | |} | | } | |_|dk rt|j||jf|dS)aSave the masks to the output path. The output is either a binary mask or a mask of objects with unique values. Args: output (str, optional): The path to the output image. Defaults to None, saving the masks to SamGeo.objects. foreground (bool, optional): Whether to generate the foreground mask. Defaults to True. unique (bool, optional): Whether to assign a unique value to each object. Defaults to True. erosion_kernel (tuple, optional): The erosion kernel for filtering object masks and extract borders. Such as (3, 3) or (5, 5). Set to None to disable it. Defaults to None. mask_multiplier (int, optional): The mask multiplier for the output mask, which is usually a binary mask [0, 1]. You can use this parameter to scale the mask to a larger range, for example [0, 255]. Defaults to 255. Nz,No masks found. Please run generate() first.r)icSs|dSNZarear$xr$r$r%z#SamGeo.save_masks..Fkeyreverserr,rr*r-)rrKrr.lenr0r2Zuint16Zuint32sortedr1 enumerater3r4r5r6rarray_to_imager)r"rRr7rSr8r9r#r:r;r<rr+Z sorted_masksrindexannr?r=r>r@rArBr$r$r%rQsN      zSamGeo.save_masks binary_roffcKsrddlm}|jr"t|j|_n|jdkr@|jfd|i||j|d|j |j|d| || dS)aShow the binary mask or the mask of objects with unique values. Args: figsize (tuple, optional): The figure size. Defaults to (12, 10). cmap (str, optional): The colormap. Defaults to "binary_r". axis (str, optional): Whether to show the axis. Defaults to "off". foreground (bool, optional): Whether to show the foreground mask only. Defaults to True. **kwargs: Other arguments for save_masks(). rNr7figsize)cmap) matplotlib.pyplotpyplotrLr5rMrrrQfigureimshowaxisshow)r"rhrirnr7r#pltr$r$r% show_masks.s    zSamGeo.show_masksffffff?cKsddlm}|j}|jdkr(tddS|dks.TrYFr,rr(ZdpidZ bbox_inchesZtightr))alpharo)rjrkrrprintr\rlrmr]ZgcaZset_autoscale_onr0r3r.Z concatenaterandomrnr4r2rZ blend_imagesr_r)r"rhrnrurRblendr#rpannsZ sorted_annsZaxZimgrar?Z color_maskarrayr$r$r% show_annsHsP          ( zSamGeo.show_annsRGBcCst|tr^|drt|}tj|s8td|d||_t |}t |t j }||_ nt|tjrlntd|jj||ddS)zSet the input image as a numpy array. Args: image (np.ndarray): The input image as a numpy array. image_format (str, optional): The image format, can be RGB or BGR. Defaults to "RGB". rDrErFz3Input image must be either a path or a numpy array.) image_formatN)rGrHrIrJrrrrKrr5rMrNrOrr0rPr! set_image)r"rr}r$r$r%r~s     zSamGeo.set_imagec Ksp|jdkrtd|dkr(|jjdd}|j||}||_t|||jfd|i||dk rlt|||ddS)a Save the predicted mask to the output path. Args: output (str): The path to the output image. index (int, optional): The index of the mask to save. Defaults to None, which will save the mask with the highest score. mask_multiplier (int, optional): The mask multiplier for the output mask, which is usually a binary mask [0, 1]. vector (str, optional): The path to the output vector file. Defaults to None. dtype (np.dtype, optional): The data type of the output image. Defaults to np.float32. simplify_tolerance (float, optional): The maximum allowed geometry displacement. The higher this value, the smaller the number of vertices in the resulting geometry. Nz1No predictions found. Please run predict() first.r)rnr+)simplify_tolerance)rrKZargmaxrrr_rraster_to_vector) r"rRr`r9r+Zvectorrr#rzr$r$r%save_predictions zSamGeo.save_predictionfloat32c Ks>t|tr8t|}|jdk r(|d}|jjj }n8t|t rpddl }| |}tj|dd}|jjj }t|trt |}t|t rt|}t|dr|j}t|dr|j}|dk r|dk rt|j||}t|trt|}|dk r$|dkr dgt|}nt|tr$|gt|}t|trpt|t|krft|dkr^|t|}ntd t|}|j}d}t|tr|dk rt|j||}t|}t|dtr|dddf}n*tj||jd }|j ||j!j"dd }n@t|tr6|dkr6t|}t|dtr6|dddf}||_#|dksrt|dksrt|d krt|dt$rt|trt|dtr|d}|%||||||\}}}n|j&|||d d\}}}||_'||_(||_)|dk r*|dkst|dts|j*|| | | f| n|j+| || | | d| r:|||fSdS)aPredict masks for the given input prompts, using the currently set image. Args: point_coords (str | dict | list | np.ndarray, optional): A Nx2 array of point prompts to the model. Each point is in (X,Y) in pixels. It can be a path to a vector file, a GeoJSON dictionary, a list of coordinates [lon, lat], or a numpy array. Defaults to None. point_labels (list | int | np.ndarray, optional): A length N array of labels for the point prompts. 1 indicates a foreground point and 0 indicates a background point. point_crs (str, optional): The coordinate reference system (CRS) of the point prompts. boxes (list | np.ndarray, optional): A length 4 array given a box prompt to the model, in XYXY format. mask_input (np.ndarray, optional): A low resolution mask input to the model, typically coming from a previous prediction iteration. Has form 1xHxW, where for SAM, H=W=256. multimask_output (bool, optional): If true, the model will return three masks. For ambiguous input prompts (such as a single click), this will often produce better masks than a single prediction. If only a single mask is needed, the model's predicted quality score can be used to select the best mask. For non-ambiguous prompts, such as multiple input prompts, multimask_output=False can give better results. return_logits (bool, optional): If true, returns un-thresholded masks logits instead of a binary mask. output (str, optional): The path to the output image. Defaults to None. index (index, optional): The index of the mask to save. Defaults to None, which will save the mask with the highest score. mask_multiplier (int, optional): The mask multiplier for the output mask, which is usually a binary mask [0, 1]. dtype (np.dtype, optional): The data type of the output image. Defaults to np.float32. return_results (bool, optional): Whether to return the predicted masks, scores, and logits. Defaults to False. Nz epsg:4326rZGeoJSON)Zdriver point_coords point_labelsrzGThe length of point_labels must be equal to the length of point_coords.r rsT)rrboxesmultimask_output) save_args),rGrHZgpd read_filecrsZto_crsZgeometryZboundsvaluestolistdictjsondumpsZvector_to_geojsonZgeojson_to_coordshasattrrrZ coords_to_xyrlistr0rzr\intrKr!Z bbox_to_xyrZtensorr transformZapply_boxes_torchrr.rfloatpredictZ predict_torchrrrrtensor_to_numpy)r"rrrZ point_crs mask_inputr return_logitsrRr`r9r+Zreturn_resultsr#gdfrZgeojsonr!Z input_boxesZcoordsrrrr$r$r%rs.                   6   zSamGeo.predictr2cCs|j}|j}|j}t|} |dkr(d}|dd|ddddf}|d}|dksbt|dkrntddStj| d|d} t t ||D]D\} \} } t | t j r| |} | | dk| d|7} q| dk|} |dk rt| ||jfd|i|n| SdS)aConvert the predicted masks from tensors to numpy arrays. Args: index (index, optional): The index of the mask to save. Defaults to None, which will save the mask with the highest score. output (str, optional): The path to the output image. Defaults to None. mask_multiplier (int, optional): The mask multiplier for the output mask, which is usually a binary mask [0, 1]. dtype (np.dtype, optional): The data type of the output image. Defaults to np.uint8. save_args (dict, optional): Optional arguments for saving the output image. Defaults to {}. Returns: np.ndarray: The predicted mask as a numpy array. NrrzNo objects found in the image.).rr*r+)rrrr0rzZsqueezer\rvZ zeros_liker^ziprGrZTensorr numpyr4r_r)r"r`rRr9r+rrrZ image_pilZimage_npZ mask_overlayiboxr@r$r$r%r_s8    zSamGeo.tensor_to_numpy SATELLITEcKst|f|||d|S)aShow the interactive map. Args: basemap (str, optional): The basemap. It can be one of the following: SATELLITE, ROADMAP, TERRAIN, HYBRID. repeat_mode (bool, optional): Whether to use the repeat mode for draw control. Defaults to True. out_dir (str, optional): The path to the output directory. Defaults to None. Returns: leafmap.Map: The map object. )basemap repeat_modeout_dir)Z sam_map_gui)r"rrrr#r$r$r%show_maps zSamGeo.show_maprr)rrrr)c Csj|jdkrtd|j}t||||\}}||_||_||}dgt|dgt|}||_||_dS)aShow a canvas to collect foreground and background points. Args: image (str | np.ndarray): The input image. fg_color (tuple, optional): The color for the foreground points. Defaults to (0, 255, 0). bg_color (tuple, optional): The color for the background points. Defaults to (0, 0, 255). radius (int, optional): The radius of the points. Defaults to 5. Returns: tuple: A tuple of two lists of foreground and background points. NzPlease run set_image() first.rr)rrK show_canvas fg_points bg_pointsr\rr) r"Zfg_colorZbg_colorZradiusrrrrrr$r$r%rs zSamGeo.show_canvascCstjrtjdS)zClear the CUDA cache.N)rr rr)r"r$r$r%clear_cuda_caches zSamGeo.clear_cuda_cachecKst||f|SN)image_to_image)r"rr#r$r$r%rszSamGeo.image_to_imagecCs(t||d|d|d|d||}|S)Nrr)Z draw_tile)r"rZpt1Zpt2Zzoomdistrr$r$r%download_tms_as_tiffs$zSamGeo.download_tms_as_tiffcKst||fd|i|dS)agSave the result to a vector file. Args: image (str): The path to the image file. output (str): The path to the vector file. simplify_tolerance (float, optional): The maximum allowed geometry displacement. The higher this value, the smaller the number of vertices in the resulting geometry. rNr)r"rrRrr#r$r$r%rs zSamGeo.raster_to_vectorcKst||fd|i|dS)alConvert a tiff file to a gpkg file. Args: tiff_path (str): The path to the tiff file. output (str): The path to the vector file. simplify_tolerance (float, optional): The maximum allowed geometry displacement. The higher this value, the smaller the number of vertices in the resulting geometry. rNrr"Z tiff_pathrRrr#r$r$r%tiff_to_vectors zSamGeo.tiff_to_vectorcKst||fd|i|dS)ajConvert a tiff file to a gpkg file. Args: tiff_path (str): The path to the tiff file. output (str): The path to the gpkg file. simplify_tolerance (float, optional): The maximum allowed geometry displacement. The higher this value, the smaller the number of vertices in the resulting geometry. rN)Zraster_to_gpkgrr$r$r% tiff_to_gpkgs zSamGeo.tiff_to_gpkgcKst||fd|i|dS)ajConvert a tiff file to a shapefile. Args: tiff_path (str): The path to the tiff file. output (str): The path to the shapefile. simplify_tolerance (float, optional): The maximum allowed geometry displacement. The higher this value, the smaller the number of vertices in the resulting geometry. rN)Z raster_to_shprr$r$r% tiff_to_shps zSamGeo.tiff_to_shpcKst||fd|i|dS)apConvert a tiff file to a GeoJSON file. Args: tiff_path (str): The path to the tiff file. output (str): The path to the GeoJSON file. simplify_tolerance (float, optional): The maximum allowed geometry displacement. The higher this value, the smaller the number of vertices in the resulting geometry. rN)Zraster_to_geojsonrr$r$r%tiff_to_geojsons zSamGeo.tiff_to_geojson)rTNNN)Tr'r))NTFNr)T)NTTNr))rbrerfT)rbrfrrNT)r|) NNNNNTFNNr)rF)rTN)rrr)N)N)N)N)N)__name__ __module__ __qualname____doc__r&rCr/rQrqr{r~r0rrrrrrrrrrrrrrr$r$r$r%rs Y / H T  G  &  :     rcsDeZdZddZfddZdfdd Zd d Zdd dZZS)SamGeoPredictorcCs$ddlm}||_||jj|_dS)Nr)ResizeLongestSide)Z!segment_anything.utils.transformsrmodelZ image_encoderZimg_sizer)r"Z sam_modelrr$r$r%r&s zSamGeoPredictor.__init__cstt||dSr)superrr~)r"r __class__r$r%r~szSamGeoPredictor.set_imageNTFc s|r|rd|_t|} t|d|d|j| } t|d|d|j| } t| d| dg} t| d| dg} t|| | }||_|d|d|_|d|d|_t ||j|j|_ t t | ||||||\}}}|||fS)Nz EPSG:4326rrrr()rZget_crsZtransform_coordsr0rzZget_pixel_coordsgeo_boxwidthheightZ set_transform geo_transformrrr)r"src_fprrrrrrrZdst_crsswneZxsZysrZiou_predictionsZ low_res_masksrr$r%rs*    zSamGeoPredictor.predictcCs(t|}t||||j|j|j|jdSr)Z get_profileZ write_rasterrrrr)r"rdst_fprZprofiler$r$r%masks_to_geotiff=sz SamGeoPredictor.masks_to_geotiffrcCst||}t|||Sr)Z get_featuresZwrite_features)r"rrZbidxrr$r$r%geotiff_to_geojsonIs  z"SamGeoPredictor.geotiff_to_geojson)NNNNNNTF)r) rrrr&r~rrr __classcell__r$r$rr%rs  r) rrr5rrr0Zsegment_anythingrrrcommonrrr$r$r$r%s