pyvale 2025.4.0__py3-none-any.whl

pyvale/core/rasternp.py

@@ -0,0 +1,604 @@
+"""
+================================================================================
+pyvale: the python validation engine
+License: MIT
+Copyright (C) 2025 The Computer Aided Validation Team
+================================================================================
+"""
+from pathlib import Path
+from multiprocessing.pool import Pool
+import numpy as np
+import numba
+from pyvale.core.cameradata import CameraData
+from pyvale.core.cameratools import CameraTools
+from pyvale.core.rendermesh import RenderMeshData
+import pyvale.core.cython.rastercyth as rastercyth
+
+
+# NOTE: This module is a feature under developement.
+
+
+class RasterNP:
+    @staticmethod
+    def world_to_raster_coords(cam_data: CameraData,
+                               coords_world: np.ndarray) -> np.ndarray:
+        # coords_world.shape=(num_nodes,coord[X,Y,Z,W])
+
+        # Project onto camera coords
+        # coords_raster.shape=(num_nodes,coord[X,Y,Z,W])
+        coords_raster = np.matmul(coords_world,cam_data.world_to_cam_mat.T)
+
+        # NOTE: w is not 1 when the matrix is a perspective projection! It is only 1
+        # here when we have an affine transformation
+        coords_raster[:,0] = coords_raster[:,0] / coords_raster[:,3]
+        coords_raster[:,1] = coords_raster[:,1] / coords_raster[:,3]
+        coords_raster[:,2] = coords_raster[:,2] / coords_raster[:,3]
+
+        # Coords Image: Perspective divide
+        coords_raster[:,0] = (cam_data.image_dist * coords_raster[:,0]
+                            / -coords_raster[:,2])
+        coords_raster[:,1] = (cam_data.image_dist * coords_raster[:,1]
+                            / -coords_raster[:,2])
+
+        # Coords NDC: Convert to normalised device coords in the range [-1,1]
+        coords_raster[:,0] = 2*coords_raster[:,0] / cam_data.image_dims[0]
+        coords_raster[:,1] = 2*coords_raster[:,1] / cam_data.image_dims[1]
+
+        # Coords Raster: Covert to pixel (raster) coords
+        # Shape = ([X,Y,Z],num_nodes)
+        coords_raster[:,0] = (coords_raster[:,0] + 1)/2 * cam_data.pixels_num[0]
+        coords_raster[:,1] = (1-coords_raster[:,1])/2 * cam_data.pixels_num[1]
+        coords_raster[:,2] = -coords_raster[:,2]
+
+        return coords_raster
+
+
+    @staticmethod
+    def back_face_removal_mask(cam_data: CameraData,
+                            coords_world: np.ndarray,
+                            connect: np.ndarray
+                            ) -> np.ndarray:
+        coords_cam = np.matmul(coords_world,cam_data.world_to_cam_mat.T)
+
+        # shape=(num_elems,nodes_per_elem,coord[x,y,z,w])
+        elem_cam_coords = coords_cam[connect,:]
+
+        # Calculate the normal vectors for all of the elements, remove the w coord
+        # shape=(num_elems,coord[x,y,z])
+        elem_cam_edge0 = elem_cam_coords[:,1,:-1] - elem_cam_coords[:,0,:-1]
+        elem_cam_edge1 = elem_cam_coords[:,2,:-1] - elem_cam_coords[:,0,:-1]
+        elem_cam_normals = np.cross(elem_cam_edge0,elem_cam_edge1,
+                                    axisa=1,axisb=1).T
+        elem_cam_normals = elem_cam_normals / np.linalg.norm(elem_cam_normals,axis=0)
+
+        cam_normal = np.array([0,0,1])
+        # shape=(num_elems,)
+        proj_elem_to_cam = np.dot(cam_normal,elem_cam_normals)
+
+        # NOTE this should be a numerical precision tolerance (epsilon)
+        back_face_mask = proj_elem_to_cam > 1e-6
+
+        return back_face_mask
+
+    @staticmethod
+    def crop_and_bound_by_connect(cam_data: CameraData,
+                                  coords_raster: np.ndarray,
+                                  connectivity: np.ndarray,
+                                  ) -> tuple[np.ndarray,np.ndarray]:
+
+        #shape=(num_elems,coord[x,y,z,w])
+        if coords_raster.ndim == 2:
+            coords_by_elem = coords_raster[connectivity,:]
+        else:
+            coords_by_elem = coords_raster[connectivity,:,:]
+
+        elem_raster_coord_min = np.min(coords_by_elem,axis=1)
+        elem_raster_coord_max = np.max(coords_by_elem,axis=1)
+
+        # Check that min/max nodes are within the 4 edges of the camera image
+        #shape=(4_edges_to_check,num_elems)
+        crop_mask = np.zeros([connectivity.shape[0],4],dtype=np.int8)
+        crop_mask[elem_raster_coord_min[:,0] <= (cam_data.pixels_num[0]-1), 0] = 1
+        crop_mask[elem_raster_coord_min[:,1] <= (cam_data.pixels_num[1]-1), 1] = 1
+        crop_mask[elem_raster_coord_max[:,0] >= 0, 2] = 1
+        crop_mask[elem_raster_coord_max[:,1] >= 0, 3] = 1
+        crop_mask = np.sum(crop_mask,axis=1) == 4
+
+        # Get only the elements that are within the FOV
+        # Mask the elem coords and the max and min elem coords for processing
+        elem_raster_coord_min = elem_raster_coord_min[crop_mask,:]
+        elem_raster_coord_max = elem_raster_coord_max[crop_mask,:]
+        num_elems_in_image = elem_raster_coord_min.shape[0]
+
+
+        # Find the indices of the bounding box that each element lies within on
+        # the image, bounded by the upper and lower edges of the image
+        elem_bound_boxes_inds = np.zeros([num_elems_in_image,4],dtype=np.int32)
+        elem_bound_boxes_inds[:,0] = RasterNP.elem_bound_box_low(
+                                            elem_raster_coord_min[:,0])
+        elem_bound_boxes_inds[:,1] = RasterNP.elem_bound_box_high(
+                                            elem_raster_coord_max[:,0],
+                                            cam_data.pixels_num[0]-1)
+        elem_bound_boxes_inds[:,2] = RasterNP.elem_bound_box_low(
+                                            elem_raster_coord_min[:,1])
+        elem_bound_boxes_inds[:,3] = RasterNP.elem_bound_box_high(
+                                            elem_raster_coord_max[:,1],
+                                            cam_data.pixels_num[1]-1)
+
+        return (crop_mask,elem_bound_boxes_inds)
+
+
+    @staticmethod
+    def elem_bound_box_low(coord_min: np.ndarray) -> np.ndarray:
+        bound_elem = np.floor(coord_min).astype(np.int32)
+        bound_low = np.zeros_like(coord_min,dtype=np.int32)
+        bound_mat = np.vstack((bound_elem,bound_low))
+        return np.max(bound_mat,axis=0)
+
+
+    @staticmethod
+    def elem_bound_box_high(coord_max: np.ndarray,image_px: int) -> np.ndarray:
+        bound_elem = np.ceil(coord_max).astype(np.int32)
+        bound_high = image_px*np.ones_like(coord_max,dtype=np.int32)
+        bound_mat = np.vstack((bound_elem,bound_high))
+        bound = np.min(bound_mat,axis=0)
+        return bound
+
+
+    @staticmethod
+    def setup_frame(cam_data: CameraData,
+                    coords_world: np.ndarray,
+                    connectivity: np.ndarray,
+                    disp_field_frame: np.ndarray | None = None,
+                    ) -> tuple[np.ndarray,np.ndarray,np.ndarray]:
+
+        connect_in_frame = np.copy(connectivity)
+        coords_deform = np.copy(coords_world)
+
+        #-----------------------------------------------------------------------
+        # DEFORM MESH WITH DISPLACEMENT
+        if disp_field_frame is not None:
+            # Exclude w coord from mesh deformation
+            coords_deform[:,:-1] = coords_deform[:,:-1] + disp_field_frame
+
+        #-----------------------------------------------------------------------
+        # Convert world coords of all elements in the scene
+        # shape=(num_nodes,coord[x,y,z,w])
+        coords_raster = RasterNP.world_to_raster_coords(cam_data,
+                                                            coords_deform)
+
+        # Convert to perspective correct hyperbolic interpolation for z interp
+        # shape=(num_nodes,coord[x,y,z,w])
+        coords_raster[:,2] = 1/coords_raster[:,2]
+        # Remove w coord
+        coords_raster = coords_raster[:,:-1]
+
+        #-----------------------------------------------------------------------
+        # BACKFACE REMOVAL
+        # shape=(num_elems,)
+        back_face_mask = RasterNP.back_face_removal_mask(cam_data,
+                                                             coords_deform,
+                                                             connect_in_frame)
+        connect_in_frame = connect_in_frame[back_face_mask,:]
+
+        #-----------------------------------------------------------------------
+        # CROPPING & BOUNDING BOX OPERATIONS
+        (crop_mask,
+         elem_bound_box_inds) = RasterNP.crop_and_bound_by_connect(
+            cam_data,
+            coords_raster,
+            connect_in_frame,
+        )
+        connect_in_frame = connect_in_frame[crop_mask,:]
+
+        #-----------------------------------------------------------------------
+        # ELEMENT AREAS FOR INTERPOLATION
+        elem_raster_coords = coords_raster[connect_in_frame,:]
+        elem_areas = edge_function_slice(elem_raster_coords[:,0,:],
+                                         elem_raster_coords[:,1,:],
+                                         elem_raster_coords[:,2,:])
+
+        return (coords_raster,connect_in_frame,elem_bound_box_inds,elem_areas)
+
+
+    @staticmethod
+    def raster_elem(cam_data: CameraData,
+                    elem_raster_coords: np.ndarray,
+                    elem_bound_box_inds: np.ndarray,
+                    elem_area: float,
+                    field_divide_z: np.ndarray
+                    ) -> tuple[np.ndarray,np.ndarray,np.ndarray,np.ndarray]:
+
+        # Create the subpixel coords inside the bounding box to test with the
+        # edge function. Use the pixel indices of the bounding box.
+        bound_subpx_x = np.arange(elem_bound_box_inds[0],
+                                  elem_bound_box_inds[1],
+                                  1/cam_data.sub_samp) + 1/(2*cam_data.sub_samp)
+        bound_subpx_y = np.arange(elem_bound_box_inds[2],
+                                  elem_bound_box_inds[3],
+                                  1/cam_data.sub_samp) + 1/(2*cam_data.sub_samp)
+        (bound_subpx_grid_x,bound_subpx_grid_y) = np.meshgrid(bound_subpx_x,
+                                                              bound_subpx_y)
+        bound_coords_grid_shape = bound_subpx_grid_x.shape
+        # shape=(coord[x,y],num_subpx_in_box)
+        bound_subpx_coords_flat = np.vstack((bound_subpx_grid_x.flatten(),
+                                             bound_subpx_grid_y.flatten()))
+
+        # Create the subpixel indices for buffer slicing later
+        subpx_inds_x = np.arange(cam_data.sub_samp*elem_bound_box_inds[0],
+                                 cam_data.sub_samp*elem_bound_box_inds[1])
+        subpx_inds_y = np.arange(cam_data.sub_samp*elem_bound_box_inds[2],
+                                 cam_data.sub_samp*elem_bound_box_inds[3])
+        (subpx_inds_grid_x,subpx_inds_grid_y) = np.meshgrid(subpx_inds_x,
+                                                            subpx_inds_y)
+
+
+        # We compute the edge function for all pixels in the box to determine if the
+        # pixel is inside the element or not
+        # NOTE: first axis of element_raster_coords is the node/vertex num.
+        # shape=(num_elems_in_bound,nodes_per_elem)
+        edge = np.zeros((3,bound_subpx_coords_flat.shape[1]),dtype=np.float64)
+        edge[0,:] = edge_function(elem_raster_coords[1,:],
+                                  elem_raster_coords[2,:],
+                                  bound_subpx_coords_flat)
+        edge[1,:] = edge_function(elem_raster_coords[2,:],
+                                  elem_raster_coords[0,:],
+                                  bound_subpx_coords_flat)
+        edge[2,:] = edge_function(elem_raster_coords[0,:],
+                                  elem_raster_coords[1,:],
+                                  bound_subpx_coords_flat)
+
+        # Now we check where the edge function is above zero for all edges
+        edge_check = np.zeros_like(edge,dtype=np.int8)
+        edge_check[edge >= 0.0] = 1
+        edge_check = np.sum(edge_check, axis=0)
+        # Create a mask with the check, TODO check the 3 here for non triangles
+        edge_mask_flat = edge_check == 3
+        edge_mask_grid = np.reshape(edge_mask_flat,bound_coords_grid_shape)
+
+        # Calculate the weights for the masked pixels
+        edge_masked = edge[:,edge_mask_flat]
+        interp_weights = edge_masked / elem_area
+
+        # Compute the depth of all pixels using hyperbolic interp
+        # NOTE: second index on raster coords is Z
+        px_coord_z = 1/(elem_raster_coords[0,2] * interp_weights[0,:]
+                      + elem_raster_coords[1,2] * interp_weights[1,:]
+                      + elem_raster_coords[2,2] * interp_weights[2,:])
+
+        field_interp = ((field_divide_z[0] * interp_weights[0,:]
+                       + field_divide_z[1] * interp_weights[1,:]
+                       + field_divide_z[2] * interp_weights[2,:])
+                       * px_coord_z)
+
+        return (px_coord_z,
+                field_interp,
+                subpx_inds_grid_x[edge_mask_grid],
+                subpx_inds_grid_y[edge_mask_grid])
+
+    @staticmethod
+    def raster_frame(cam_data: CameraData,
+                     connect_in_frame: np.ndarray,
+                     coords_raster: np.ndarray,
+                     elem_bound_box_inds: np.ndarray,
+                     elem_areas: np.ndarray,
+                     field_frame_div_z: np.ndarray
+                     ) -> tuple[np.ndarray,np.ndarray,int]:
+        #connect_in_frame.shape=(num_elems,nodes_per_elem)
+        #coords_raster.shape=(num_coords,coord[x,y,z,w])
+        #elem_bound_box_inds.shape=(num_elems,[min_x,max_x,min_y,max_y])
+        #elem_areas.shape=(num_elems,)
+        #field_frame_divide_z=(num_coords,)
+
+        depth_buffer = 1e5*cam_data.image_dist*np.ones(
+            cam_data.sub_samp*cam_data.pixels_num).T
+        image_buffer = np.full(cam_data.sub_samp*cam_data.pixels_num,0.0).T
+
+        # elem_raster_coords.shape=(num_elems,nodes_per_elem,coord[x,y,z,w])
+        # field_divide_z.shape=(num_elems,nodes_per_elem,num_time_steps)
+        # elem_raster_coords.shape=(nodes_per_elem,coord[x,y,z,w])
+
+        for ee in range(connect_in_frame.shape[0]):
+            cc = connect_in_frame[ee,:]
+
+            (px_coord_z,
+            field_interp,
+            subpx_inds_x_in,
+            subpx_inds_y_in) = RasterNP.raster_elem(
+                                                cam_data,
+                                                coords_raster[cc,:],
+                                                elem_bound_box_inds[ee,:],
+                                                elem_areas[ee],
+                                                field_frame_div_z[cc])
+
+
+            #  Build a mask to replace the depth information if there is already an
+            # element in front of the one we are rendering
+            px_coord_z_depth_mask = (px_coord_z
+                < depth_buffer[subpx_inds_y_in,subpx_inds_x_in])
+
+            # Initialise the z coord to the value in the depth buffer
+            px_coord_z_masked = depth_buffer[subpx_inds_y_in,subpx_inds_x_in]
+            # Use the depth mask to overwrite the depth buffer values if points are in
+            # front of the values in the depth buffer
+            px_coord_z_masked[px_coord_z_depth_mask] = px_coord_z[px_coord_z_depth_mask]
+
+            # Push the masked values into the depth buffer
+            depth_buffer[subpx_inds_y_in,subpx_inds_x_in] = px_coord_z_masked
+
+            # Mask the image buffer using the depth mask
+            image_buffer_depth_masked = image_buffer[subpx_inds_y_in,subpx_inds_x_in]
+            image_buffer_depth_masked[px_coord_z_depth_mask] = field_interp[px_coord_z_depth_mask]
+
+            # Push the masked values into the image buffer
+            image_buffer[subpx_inds_y_in,subpx_inds_x_in] = image_buffer_depth_masked
+
+        #---------------------------------------------------------------------------
+        # END RASTER LOOP
+        # TODO: fix this for windows
+        if Path(rastercyth.__file__).suffix == ".so":
+            depth_buff = np.empty((cam_data.pixels_num[1],cam_data.pixels_num[0]),dtype=np.float64)
+            depth_buff = np.array(rastercyth.average_image(depth_buffer,cam_data.sub_samp))
+            image_buff = np.empty((cam_data.pixels_num[1],cam_data.pixels_num[0]),dtype=np.float64)
+            image_buff = np.array(rastercyth.average_image(image_buffer,cam_data.sub_samp))
+        else:
+            depth_buff = CameraTools.average_subpixel_image(depth_buffer,cam_data.sub_samp)
+            image_buff = CameraTools.average_subpixel_image(image_buffer,cam_data.sub_samp)
+
+        return (image_buff,depth_buff)
+
+
+    @staticmethod
+    def raster_static_mesh(cam_data: CameraData,
+                           render_mesh: RenderMeshData,
+                           save_path: Path | None = None,
+                           threads_num: int | None = None,
+                           ) -> np.ndarray | None:
+
+        frames_num = render_mesh.fields_render.shape[1]
+        field_num = render_mesh.fields_render.shape[2]
+        (frames,fields) = np.meshgrid(np.arange(0,frames_num),
+                                       np.arange(0,field_num))
+        frames = frames.flatten()
+        fields = fields.flatten()
+
+        if save_path is None:
+            images = np.empty((cam_data.pixels_num[1],
+                            cam_data.pixels_num[0],
+                            frames_num,
+                            field_num))
+        else:
+            images = None
+            if not save_path.is_dir():
+                save_path.mkdir()
+
+        # DO THIS ONCE: for non deforming meshes
+        # coords_raster.shape=(num_coords,coord[x,y,z,w])
+        # connect_in_frame.shape=(num_elems_in_scene,nodes_per_elem)
+        # elem_bound_box_inds.shape=(num_elems_in_scene,4[x_min,x_max,y_min,y_max])
+        # elem_areas.shape=(num_elems,)
+        (coords_raster,
+        connect_in_frame,
+        elem_bound_box_inds,
+        elem_areas) = RasterNP.setup_frame(
+            cam_data,
+            render_mesh.coords,
+            render_mesh.connectivity,
+        )
+
+        if threads_num is None:
+            for ff in range(0,frames.shape[0]):
+                image = RasterNP._static_mesh_frame_loop(
+                    frames[ff],
+                    fields[ff],
+                    cam_data,
+                    coords_raster,
+                    connect_in_frame,
+                    elem_bound_box_inds,
+                    elem_areas,
+                    render_mesh.fields_render[:,frames[ff],fields[ff]],
+                    save_path
+                )
+
+                if images is not None:
+                    images[:,:,frames[ff],fields[ff]] = image
+        else:
+            with Pool(threads_num) as pool:
+                processes_with_id = []
+
+                for ff in range(0,frames.shape[0]):
+                    args = (frames[ff],
+                            fields[ff],
+                            cam_data,
+                            coords_raster,
+                            connect_in_frame,
+                            elem_bound_box_inds,
+                            elem_areas,
+                            render_mesh.fields_render[:,frames[ff],fields[ff]],
+                            save_path)
+
+                    process = pool.apply_async(
+                            RasterNP._static_mesh_frame_loop, args=args
+                    )
+                    processes_with_id.append({"process": process,
+                                              "frame": frames[ff],
+                                              "field": fields[ff]})
+
+                for pp in processes_with_id:
+                    image = pp["process"].get()
+                    images[:,:,pp["frame"],pp["field"]] = image
+
+        if images is not None:
+            return images
+
+        return None
+
+
+
+
+    @staticmethod
+    def raster_deformed_mesh(cam_data: CameraData,
+                             render_mesh: RenderMeshData,
+                             save_path: Path | None = None,
+                             parallel: int | None = None
+                             ) -> np.ndarray | None:
+
+        frames_num = render_mesh.fields_render.shape[1]
+        field_num = render_mesh.fields_render.shape[2]
+        (frames,fields) = np.meshgrid(np.arange(0,frames_num),
+                                       np.arange(0,field_num))
+        frames = frames.flatten()
+        fields = fields.flatten()
+
+
+        if save_path is None:
+            images = np.empty((cam_data.pixels_num[1],
+                               cam_data.pixels_num[0],
+                               frames_num,
+                               field_num))
+        else:
+            images = None
+            if not save_path.is_dir():
+                save_path.mkdir()
+
+
+        if parallel is None:
+            for ff in range(0,frames.shape[0]):
+                image = RasterNP._deformed_mesh_frame_loop(
+                    frames[ff],
+                    fields[ff],
+                    cam_data,
+                    render_mesh,
+                    save_path,
+                )
+
+                if images is not None:
+                    images[:,:,frames[ff],fields[ff]] = image
+        else:
+            with Pool(parallel) as pool:
+                processes_with_id = []
+
+                for ff in range(0,frames.shape[0]):
+                    args = (frames[ff],
+                            fields[ff],
+                            cam_data,
+                            render_mesh,
+                            save_path)
+
+                    process = pool.apply_async(
+                            RasterNP._deformed_mesh_frame_loop, args=args
+                    )
+                    processes_with_id.append({"process": process,
+                                              "frame": frames[ff],
+                                              "field": fields[ff]})
+
+                for pp in processes_with_id:
+                    image = pp["process"].get()
+                    images[:,:,pp["frame"],pp["field"]] = image
+
+        if images is not None:
+            return images
+
+        return None
+
+
+    @staticmethod
+    def _static_mesh_frame_loop(frame_ind: int,
+                                field_ind: int,
+                                cam_data: CameraData,
+                                coords_raster: np.ndarray,
+                                connect_in_frame: np.ndarray,
+                                elem_bound_box_inds: np.ndarray,
+                                elem_areas: np.ndarray,
+                                field_to_render: np.ndarray,
+                                save_path: Path | None,
+                                ) -> np.ndarray | None:
+
+        # NOTE: the z coord has already been inverted in setup so we multiply here
+        render_field_div_z = field_to_render*coords_raster[:,2]
+
+        (image_buffer,
+        depth_buffer) = RasterNP.raster_frame(
+                                    cam_data,
+                                    connect_in_frame,
+                                    coords_raster,
+                                    elem_bound_box_inds,
+                                    elem_areas,
+                                    render_field_div_z)
+
+        # TODO: make this configurable
+        image_buffer[depth_buffer > 1000*cam_data.image_dist] = np.nan
+
+        if save_path is None:
+            return image_buffer
+
+        image_file = save_path/f"image_frame{frame_ind}_field{field_ind}"
+        np.save(image_file,image_buffer)
+        return None
+
+
+    @staticmethod
+    def _deformed_mesh_frame_loop(frame_ind: int,
+                                  field_ind: int,
+                                  cam_data: CameraData,
+                                  render_mesh: RenderMeshData,
+                                  save_path: Path | None
+                                  ) -> np.ndarray | None:
+        # coords_raster.shape=(num_coords,coord[x,y,z,w])
+        # connect_in_frame.shape=(num_elems_in_scene,nodes_per_elem)
+        # elem_bound_box_inds.shape=(num_elems_in_scene,4[x_min,x_max,y_min,y_max])
+        # elem_areas.shape=(num_elems,)
+        (coords_raster,
+        connect_in_frame,
+        elem_bound_box_inds,
+        elem_areas) = RasterNP.setup_frame(
+            cam_data,
+            render_mesh.coords,
+            render_mesh.connectivity,
+            render_mesh.fields_disp[:,frame_ind,:],
+        )
+
+        # NOTE: the z coord has already been inverted in setup so we multiply here
+        render_field_div_z = (render_mesh.fields_render[:,frame_ind,field_ind]
+                            *coords_raster[:,2])
+
+        # image_buffer.shape=(num_px_y,num_px_x)
+        # depth_buffer.shape=(num_px_y,num_px_x)
+        (image_buffer,
+        depth_buffer) = RasterNP.raster_frame(
+                                    cam_data,
+                                    connect_in_frame,
+                                    coords_raster,
+                                    elem_bound_box_inds,
+                                    elem_areas,
+                                    render_field_div_z)
+
+        # TODO: make this configurable
+        image_buffer[depth_buffer > 1000*cam_data.image_dist] = np.nan
+
+        if save_path is None:
+            return image_buffer
+
+        image_file = save_path/f"image_frame{frame_ind}_field{field_ind}"
+        np.save(image_file.with_suffix(".npy"),image_buffer)
+        return None
+
+
+#-------------------------------------------------------------------------------
+@numba.jit(nopython=True)
+def edge_function(vert_a: np.ndarray,
+                  vert_b: np.ndarray,
+                  vert_c: np.ndarray) -> np.ndarray:
+
+    return  ((vert_c[0] - vert_a[0]) * (vert_b[1] - vert_a[1])
+              - (vert_c[1] - vert_a[1]) * (vert_b[0] - vert_a[0]))
+
+@numba.jit(nopython=True)
+def edge_function_slice(vert_a: np.ndarray,
+                        vert_b: np.ndarray,
+                        vert_c: np.ndarray) -> np.ndarray:
+
+    return  ((vert_c[:,0] - vert_a[:,0]) * (vert_b[:,1] - vert_a[:,1])
+              - (vert_c[:,1] - vert_a[:,1]) * (vert_b[:,0] - vert_a[:,0]))
+
+