| # Where to save the downloaded dataset | |
| dataset_dir: "datasets/new_locations" | |
| fpv_options: | |
| # Pipeline configuration to filter FPV images | |
| filter_pipeline_cfg: "mia/fpv/filter_pipelines/mia.yaml" | |
| # Chunk size is used for checkpointing as image metadata | |
| # can be very memory intensive. This allows you to resume | |
| # from the last chunk in case things go wrong. | |
| metadata_download_chunk_size: 50000 | |
| # FPV retrieval is comprised of the stages bellow | |
| # these boolean flags allow you to only execute certain stages | |
| # Note that the stages are ordered and later stages assume previous stages | |
| # are complete. | |
| stages: | |
| get_image_points_from_tiles: True | |
| get_metadata: True | |
| run_filter: True | |
| download_images: True | |
| to_process_sequence: True | |
| bev_options: | |
| # Local planet dump OSM file path. File format should be either .osm or .json. | |
| # If not provided, tiled OSM data will be downloaded from the internet. | |
| # Additionally, the OSM data will be used to clip images such that they all lie within its boundary box. | |
| osm_fp: | |
| # Download the OSM file encompoassing the whole area + some padding | |
| # Pretty slow to render but can fix the problem of missing road segments, | |
| # which happens when the small bboxes do not contain any of the ends of the road segment. | |
| one_big_osm: False | |
| # Padding in meters that allows rendering a bigger bounding box then cropping. | |
| # Useful to reduce the problem of missing road segments. | |
| # If one_big_osm option is turned on, this padding is added over the big map only. | |
| padding: 50 | |
| # Download OSM data only and do not process maps into semantic masks. | |
| download_osm_only: False | |
| # If enabled, the osm_cache will store a file per ID which removes the need for synchronization | |
| # between processes. If disabled, osm_cache will store files based on bbox queries, enabling | |
| # reuse of osm tiles. This was observed to reduce needed OSM downloads by ~20% but will | |
| # trigger file lock synchronization if using multiworkers to avoid race conditions. | |
| store_osm_per_id: False | |
| # MapMachine style sheet | |
| map_machine_scheme: "mia/bev/styles/mia.yml" | |
| #Final map pixel size after rotation | |
| map_length: 224 | |
| # Map resolution in meters per pixel | |
| meters_per_pixel: 0.5 | |
| # If downsampling the map after processing is desired | |
| # You can use the below downsampling factor | |
| final_downsample: 1 | |
| # Store satelite images as well using google earth engine. | |
| # Requires you to already have a google earth engine project, be authenticated using `earthengine authenticate` | |
| # and a the project id set using gcloud auth `gcloud auth application-default set-quota-project PROJECT_ID`) | |
| store_sat: False | |
| # Whether or not to store RAW BEV svgs and rendered semantic masks. | |
| store_all_steps: False | |
| # How many processes to use to process images. Set to 0 to disable multiprocessing. | |
| n_workers: 0 | |
| # Redownload existing BEV images. Useful if style sheet is updated. | |
| redownload: False | |
| # Should we sample from the dataframe instead of downloading everything? | |
| # Set to -1 to download everything | |
| n_samples: -1 | |
| # List all locations you want to download and process below | |
| cities: | |
| - name: "Mount Oliver" | |
| state: "Pennsylvania" # (Optional) | |
| country: "United States" # (Optional) | |
| bound_type: "auto_shape" # ["auto_shape", "auto_bbox", "custom_size", "custom_bbox"] | |
| - name: "Greensburg" | |
| state: "Pennsylvania" | |
| bound_type: "custom_size" | |
| custom_size: 5 # 5km x 5km centered on auto-fetched city center | |
| - name: "Frick Park" | |
| bound_type: "custom_bbox" | |
| custom_bbox: "-79.9086,40.4294,-79.9052,40.4315" # East,South,West,North |