cars.applications.dense_matching.dense_matching_wrappers

This module is responsible for the dense matching algorithms: - thus it creates a disparity map from a pair of images

Functions

get_margins(margin, disp_min, disp_max)	Get margins for the dense matching steps
get_masks_from_pandora(→ Dict[str, numpy.ndarray])	Get masks dictionary from the disparity map in output of pandora.
add_disparity_grids(output_dataset[, disp_min_grid, ...])	Add disparity min and max grids to dataset
add_texture(output_dataset[, texture, color_type, ...])	Add image and image mask to dataset
add_edges(output_dataset, ref_dataset)	Add image and image mask to dataset
add_classification(output_dataset[, classif, band_classif])	Add classification to dataset
compute_cropped_roi(current_margins, margins_to_keep, ...)	Compute cropped roi, with associated margins and
create_disp_dataset(→ xarray.Dataset)	Create the disparity dataset.
add_crop_info(disp_ds, cropped_range)	Add crop info
estimate_right_classif_on_left(right_classif, ...)	Estimate right classif on left image
mask_left_classif_from_right_mask(left_classif, ...)	Mask left classif with right mask.
merge_classif_left_right(left_classif, ...)	Merge left and right classif
add_confidence(output_dataset, disp)	Add confidences to dataset
to_safe_disp_grid(grid_disp_min, grid_disp_max)	Generate safe grids, with min < max for each point
estimate_right_grid_disp(disp_min_grid, disp_max_grid)	Estimate right grid min and max.
optimal_tile_size_pandora_plugin_libsgm(→ int)	Compute optimal tile size according to estimated memory usage
get_max_disp_from_opt_tile_size(...[, margin, ...])	Compute max range possible depending on max ram per worker
nan_ratio_func(window)
confidence_filtering(dataset, requested_confidence, ...)	Filter the disparity map by using the confidence

Module Contents

cars.applications.dense_matching.dense_matching_wrappers.get_margins(margin, disp_min, disp_max)

Get margins for the dense matching steps

Parameters:

• margin (Margins) – margins object

• disp_min (int) – Minimum disparity

• disp_max (int) – Maximum disparity

Returns:

margins of the matching algorithm used

cars.applications.dense_matching.dense_matching_wrappers.get_masks_from_pandora(disp: xarray.Dataset, compute_disparity_masks: bool, filter_incomplete_disparity_range: bool) → Dict[str, numpy.ndarray]

Get masks dictionary from the disparity map in output of pandora.

Parameters:

• disp – disparity map (pandora output)

• compute_disparity_masks – compute_disparity_masks activation status

Returns:

masks dictionary

cars.applications.dense_matching.dense_matching_wrappers.add_disparity_grids(output_dataset: xarray.Dataset, disp_min_grid: numpy.ndarray = None, disp_max_grid: numpy.ndarray = None)

Add disparity min and max grids to dataset

Parameters:

• output_dataset – output dataset

• disp_min_grid – dense disparity map grid min

• disp_max_grid – dense disparity map grid max

cars.applications.dense_matching.dense_matching_wrappers.add_texture(output_dataset: xarray.Dataset, texture: numpy.ndarray = None, color_type=None, band_im: list = None, texture_bands: list = None)

Add image and image mask to dataset

Parameters:

• output_dataset – output dataset

• image – image array

• image_type – data type of pixels

• band_im – list of band names

cars.applications.dense_matching.dense_matching_wrappers.add_edges(output_dataset: xarray.Dataset, ref_dataset: xarray.Dataset)

Add image and image mask to dataset

Parameters:

• output_dataset – output dataset

• image – image array

• image_type – data type of pixels

• band_im – list of band names

cars.applications.dense_matching.dense_matching_wrappers.add_classification(output_dataset: xarray.Dataset, classif: numpy.ndarray = None, band_classif: list = None)

Add classification to dataset

Parameters:

• output_dataset – output dataset

• classif – classif array

• band_im – list of band names

cars.applications.dense_matching.dense_matching_wrappers.compute_cropped_roi(current_margins, margins_to_keep, tile_roi, nb_rows, nb_cols)

Compute cropped roi, with associated margins and

Parameters:

• current_margins (list[int]) – current dataset margins

• margins_to_keep (int) – margin to keep

• nb_rows – number of current rows

• tile_roi (list) – roi without margin of tile

• nb_cols (int) – number of current cols

Returns:

(borders to use as roi, new dataset roi with margin, margin associated to roi)

Return type:

tuple(list, list, list)

cars.applications.dense_matching.dense_matching_wrappers.create_disp_dataset(disp: xarray.Dataset, ref_dataset: xarray.Dataset, secondary_dataset: xarray.Dataset, compute_disparity_masks: bool = False, disp_min_grid=None, disp_max_grid=None, cropped_range=None, margins_to_keep=0, classification_fusion_margin=-1, texture_bands=None, filter_incomplete_disparity_range=True) → xarray.Dataset

Create the disparity dataset.

Parameters:

• disp – disparity map (result of pandora)

• ref_dataset – reference dataset for the considered disparity map

• secondary_dataset – secondary dataset for the considered disparity map

• compute_disparity_masks – compute_disparity_masks activation status

• disp_min_grid – disparity min grid

• disp_max_grid – disparity max grid

• cropped_range (numpy array) – true if disparity range was cropped

• margins_to_keep (int) – margin to keep after dense matching

Returns:

disparity dataset as used in cars

cars.applications.dense_matching.dense_matching_wrappers.add_crop_info(disp_ds, cropped_range)

Add crop info

Parameters:

• disp – disp xarray

• cropped_range – was cropped range, bool

:return updated dataset

cars.applications.dense_matching.dense_matching_wrappers.estimate_right_classif_on_left(right_classif, disp_map, disp_mask, disp_min, disp_max, classifiation_fusion_margin)

Estimate right classif on left image

Parameters:

• right_classif (np ndarray) – right classification

• disp_map (np ndarray) – disparity map

• disp_mask (np ndarray) – disparity mask

• disp_min (int) – disparity min

• disp_max (int) – disparity max

Returns:

right classif on left image

Return type:

np nadarray

cars.applications.dense_matching.dense_matching_wrappers.mask_left_classif_from_right_mask(left_classif, right_mask, disp_min, disp_max, classifiation_fusion_margin)

Mask left classif with right mask.

Parameters:

• left_classif (np ndarray) – right classification

• right_mask (np ndarray) – right mask

• disp_min (np.array type int) – disparity min

• disp_max (np.array type int) – disparity max

Returns:

masked left classif

Return type:

np nadarray

cars.applications.dense_matching.dense_matching_wrappers.merge_classif_left_right(left_classif, left_band_classif, right_classif, right_band_classif)

Merge left and right classif

Parameters:

• left_classif (np nadarray) – left classif

• left_band_classif (list) – list of tag

• right_classif (np nadarray) – left classif

• right_band_classif (list) – list of tag

Returns:

merged classif, merged tag list

Return type:

np ndarray, list

cars.applications.dense_matching.dense_matching_wrappers.add_confidence(output_dataset: xarray.Dataset, disp: xarray.Dataset)

Add confidences to dataset

Parameters:

• output_dataset – output dataset

• disp – disp xarray

cars.applications.dense_matching.dense_matching_wrappers.to_safe_disp_grid(grid_disp_min, grid_disp_max)

Generate safe grids, with min < max for each point

Parameters:

• grid_disp_min – min disp grid

• grid_disp_max – max disp grid

Returns:

grid_disp_min, grid_disp_max

cars.applications.dense_matching.dense_matching_wrappers.estimate_right_grid_disp(disp_min_grid, disp_max_grid)

Estimate right grid min and max. Correspond to the range of pixels that can be correlated from left -> right. If no left pixels can be associated to right, use global values

Parameters:

• disp_min_grid (numpy ndarray) – left disp min grid

• disp_max_grid (numpy ndarray) – left disp max grid

Returns:

disp_min_right_grid, disp_max_right_grid

Return type:

numpy ndarray, numpy ndarray

cars.applications.dense_matching.dense_matching_wrappers.optimal_tile_size_pandora_plugin_libsgm(disp_min: int, disp_max: int, min_tile_size: int, max_tile_size: int, max_ram_per_worker: int, tile_size_rounding: int = 50, margin: int = 0) → int

Compute optimal tile size according to estimated memory usage (pandora_plugin_libsgm) Returned optimal tile size will be at least equal to tile_size_rounding.

Parameters:

• disp_min – Minimum disparity to explore

• disp_max – Maximum disparity to explore

• min_tile_size – Minimal tile size

• max_tile_size – Maximal tile size

• max_ram_per_worker – amount of RAM allocated per worker

• tile_size_rounding – Optimal tile size will be aligned to multiples of tile_size_rounding

• margin – margin to remove to the computed tile size (as a percent of the computed tile size)

Returns:

Optimal tile size according to benchmarked memory usage

cars.applications.dense_matching.dense_matching_wrappers.get_max_disp_from_opt_tile_size(opt_epipolar_tile_size, max_ram_per_worker, margin=0, used_disparity_range=0)

Compute max range possible depending on max ram per worker Return max range usable

Parameters:

• opt_epipolar_tile_size – used tile size

• max_ram_per_worker – amount of RAM allocated per worker

• tile_size_rounding – Optimal tile size will be aligned to multiples of tile_size_rounding

• margin – margin to remove to the computed tile size (as a percent of the computed tile size)

Returns:

max disp range to use

cars.applications.dense_matching.dense_matching_wrappers.nan_ratio_func(window)

” Calculate the number of nan in the window

Parameters:

window – the window in the image

cars.applications.dense_matching.dense_matching_wrappers.confidence_filtering(dataset, requested_confidence, conf_filtering)

Filter the disparity map by using the confidence

Parameters:

• dataset (cars dataset) – the epipolar disparity map dataset

• disp_map (numpy darray) – the disparity map

• requested_confidence (list) – the confidence to use

• conf_filtering (dict) – the confidence_filtering parameters