Add support for the Eiffel Tower underwater dataset

hloc/pipelines/EiffelTower/README.md

@@ -0,0 +1,24 @@
+# Eiffel Tower underwater dataset
+
+## Installation
+
+Download the dataset from the [SEANOE](https://www.seanoe.org/data/00810/92226/):
+```bash
+export dataset=datasets/eiffeltower
+export IDs=( "98240" "98289" "98314" "98356" "98357" )
+for id in "${IDs[@]}"; do
+wget https://www.seanoe.org/data/00810/92226/data/$id.zip -P $dataset \
+&& unzip $dataset/$id.zip -d $dataset && rm $dataset/$id.zip; done
+mkdir $dataset/global/images
+for year in 2015 2016 2018 2020; do
+for image in `ls $dataset/$year/images/`; do
+ln -s ../../$year/images/$image $dataset/global/images/;
+done; done
+```
+
+
+## Pipeline
+
+```bash
+python3 -m hloc.pipelines.EiffelTower.pipeline
+```

hloc/pipelines/EiffelTower/pipeline.py

@@ -0,0 +1,114 @@
+import argparse
+from pathlib import Path
+from pprint import pformat
+
+import pycolmap
+
+from ..Cambridge.utils import create_query_list_with_intrinsics, evaluate
+from ... import extract_features, match_features, triangulation
+from ... import logger
+from ... import pairs_from_covisibility, pairs_from_retrieval, localize_sfm
+from ...utils.parsers import parse_image_list
+
+
+def create_query_list(model: Path, queries: Path):
+    reconstruction = pycolmap.Reconstruction(model)
+    queries.parent.mkdir(parents=True, exist_ok=True)
+    with open(queries, 'w') as q:
+        for image in reconstruction.images.values():
+            if image.name[:4] == '2015':  # images of the 2015 visit are the query images
+                q.write(f'{image.name}\n')
+
+
+def create_db_sfm(model: Path, queries: Path, output: Path):
+    '''Filter the model to discard the query images and keep only the reference images.
+    '''
+    reconstruction = pycolmap.Reconstruction(model)
+    query_names = parse_image_list(queries)
+    query_ids = [image.image_id for image in reconstruction.images.values() if image.name in query_names]
+    for query_id in query_ids:
+        reconstruction.deregister_image(query_id)
+    output.mkdir(parents=True, exist_ok=True)
+    reconstruction.write(output)
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--dataset', type=Path, default='datasets/eiffeltower',
+                        help='Path to the dataset, default: %(default)s')
+    parser.add_argument('--outputs', type=Path, default='outputs/eiffeltower',
+                        help='Path to the output directory, default: %(default)s')
+    parser.add_argument('--num_covis', type=int, default=20,
+                        help='Number of image pairs for SfM, default: %(default)s')
+    parser.add_argument('--num_loc', type=int, default=10,
+                        help='Number of image pairs for loc, default: %(default)s')
+    args = parser.parse_args()
+
+    # set up the paths
+    dataset = args.dataset
+    images = dataset / 'global' / 'images'
+    global_sfm = dataset / 'global' / 'sfm'  # the global SfM model with both reference and query images
+
+    outputs = args.outputs  # where everything will be saved
+    query_list = outputs / 'list_query.txt'  # where to save the list of query images
+    query_list_with_intrinsics = outputs / 'list_query_with_intrinsics.txt'  # list of query images with intrinsics
+    db_sfm = outputs / 'sfm_db'  # where to save the global SfM filtered with only the reference images
+    reference_sfm = outputs / 'sfm_superpoint+superglue'  # the SfM model we will build
+    sfm_pairs = outputs / f'pairs-db-covis{args.num_covis}.txt'  # top-k most covisible in dataset model
+    loc_pairs = outputs / f'pairs-query-netvlad{args.num_loc}.txt'  # top-k retrieved by NetVLAD
+    results = outputs / f'EiffelTower_hloc_superpoint+superglue_netvlad{args.num_loc}.txt'
+
+    # list the standard configurations available
+    logger.info(f'Configs for feature extractors:\n{pformat(extract_features.confs)}')
+    logger.info(f'Configs for feature matchers:\n{pformat(match_features.confs)}')
+
+    # pick one of the configurations for extraction and matching
+    retrieval_conf = extract_features.confs['netvlad']
+    feature_conf = extract_features.confs['superpoint_aachen']
+    matcher_conf = match_features.confs['superglue']
+
+    # set up query lists and database model from the Eiffel Tower dataset
+    logger.info(f'Create lists of query images: {query_list}.')
+    create_query_list(global_sfm, query_list)
+
+    logger.info(f'Create list of query images with intrinsics: {query_list_with_intrinsics}.')
+    create_query_list_with_intrinsics(global_sfm, query_list_with_intrinsics, list_file=query_list, ext='.txt')
+
+    logger.info(f'Filter global SfM model to keep only the reference images: {db_sfm}.')
+    create_db_sfm(global_sfm, query_list, db_sfm)
+
+    features = extract_features.main(feature_conf, images, outputs)
+
+    pairs_from_covisibility.main(
+        db_sfm, sfm_pairs, num_matched=args.num_covis)
+    sfm_matches = match_features.main(
+        matcher_conf, sfm_pairs, feature_conf['output'], outputs)
+
+    triangulation.main(
+        reference_sfm,
+        db_sfm,
+        images,
+        sfm_pairs,
+        features,
+        sfm_matches)
+
+    global_descriptors = extract_features.main(retrieval_conf, images, outputs)
+
+    pairs_from_retrieval.main(
+        global_descriptors, loc_pairs, args.num_loc,
+        query_list=query_list, db_model=reference_sfm)
+    loc_matches = match_features.main(
+        matcher_conf, loc_pairs, feature_conf['output'], outputs)
+
+    localize_sfm.main(
+        reference_sfm,
+        query_list_with_intrinsics,
+        loc_pairs,
+        features,
+        loc_matches,
+        results,
+        covisibility_clustering=False)  # not required with SuperPoint+SuperGlue
+
+    evaluate(
+        global_sfm, results,
+        query_list, ext='.txt')