Datasets:
Sub-tasks:
multi-class-image-classification
Languages:
English
Size:
100K<n<1M
ArXiv:
Tags:
Place Recognition
License:
import os | |
import pandas as pd | |
from pathlib import Path | |
import datasets | |
_CITATION = """ | |
@inproceedings{your_neurips_submission, | |
title={Multimodal Street-level Place Recognition Dataset}, | |
author={Ou, Yiwei}, | |
year={2025}, | |
booktitle={NeurIPS Datasets and Benchmarks Track} | |
} | |
""" | |
_DESCRIPTION = """ | |
Multimodal Street-level Place Recognition Dataset (Resized version). | |
This version loads images, videos, and associated annotations for place recognition tasks, | |
including GPS, camera metadata, and temporal information. | |
""" | |
_HOMEPAGE = "https://huggingface.co/datasets/Yiwei-Ou/Multimodal_Street-level_Place_Recognition_Dataset" | |
_LICENSE = "cc-by-4.0" | |
class MultimodalPlaceRecognition(datasets.GeneratorBasedBuilder): | |
VERSION = datasets.Version("1.0.0") | |
def _info(self): | |
return datasets.DatasetInfo( | |
description=_DESCRIPTION, | |
features=datasets.Features({ | |
"image_path": datasets.Value("string"), | |
"video_path": datasets.Value("string"), | |
"location_code": datasets.Value("string"), | |
"spatial_type": datasets.Value("string"), | |
"index": datasets.Value("int32"), | |
"shop_names": datasets.Value("string"), | |
"sign_text": datasets.Value("string"), | |
"image_metadata": datasets.Value("string"), | |
"video_metadata": datasets.Value("string"), | |
}), | |
supervised_keys=None, | |
homepage=_HOMEPAGE, | |
license=_LICENSE, | |
citation=_CITATION, | |
) | |
def _split_generators(self, dl_manager): | |
archive_path = dl_manager.download_and_extract( | |
"https://huggingface.co/datasets/Yiwei-Ou/Multimodal_Street-level_Place_Recognition_Dataset/resolve/main/Annotated_Resized.tar.gz" | |
) | |
base_dir = os.path.join(archive_path, "03 Annotated_Resized", "Dataset_Full") | |
image_dir = os.path.join(base_dir, "Images") | |
video_dir = os.path.join(base_dir, "Videos") | |
text_dir = os.path.join(base_dir, "Texts") | |
return [ | |
datasets.SplitGenerator( | |
name=datasets.Split.TRAIN, | |
gen_kwargs={ | |
"image_dir": image_dir, | |
"video_dir": video_dir, | |
"annotations_path": os.path.join(text_dir, "Annotations.xlsx"), | |
"image_meta_path": os.path.join(text_dir, "Media_Metadata-Images.xlsx"), | |
"video_meta_path": os.path.join(text_dir, "Media_Metadata-Videos.xlsx"), | |
}, | |
) | |
] | |
def _generate_examples(self, image_dir, video_dir, annotations_path, image_meta_path, video_meta_path): | |
id_ = 0 | |
annotations_df = pd.read_excel(annotations_path, engine="openpyxl") | |
annotations_dict = { | |
str(row["Code"]).strip(): { | |
"spatial_type": str(row["Type"]).strip(), | |
"index": int(row["Index"]), | |
"shop_names": str(row["List of Store Names and Signs"]) if not pd.isna(row["List of Store Names and Signs"]) else "", | |
"sign_text": "", # Not explicitly available | |
} | |
for _, row in annotations_df.iterrows() | |
} | |
image_meta_df = pd.read_excel(image_meta_path, engine="openpyxl") | |
image_meta_dict = { | |
str(row["Filename"]).strip(): row.drop("Filename").dropna().to_dict() | |
for _, row in image_meta_df.iterrows() | |
} | |
video_meta_df = pd.read_excel(video_meta_path, engine="openpyxl") | |
video_meta_dict = { | |
str(row["Filename"]).strip(): row.drop("Filename").dropna().to_dict() | |
for _, row in video_meta_df.iterrows() | |
} | |
for spatial_type in os.listdir(image_dir): | |
spatial_path = os.path.join(image_dir, spatial_type) | |
if not os.path.isdir(spatial_path): | |
continue | |
for location_code in os.listdir(spatial_path): | |
loc_img_path = os.path.join(spatial_path, location_code) | |
if not os.path.isdir(loc_img_path): | |
continue | |
loc_vid_path = os.path.join(video_dir, spatial_type, location_code) if os.path.exists(os.path.join(video_dir, spatial_type, location_code)) else None | |
vid_files = set(os.listdir(loc_vid_path)) if loc_vid_path else set() | |
for file_name in os.listdir(loc_img_path): | |
if file_name.lower().endswith((".jpg", ".jpeg", ".png")): | |
base_name = os.path.splitext(file_name)[0] | |
video_match = [v for v in vid_files if v.startswith(base_name) and v.endswith(".mp4")] | |
video_file = video_match[0] if video_match else "" | |
video_path = os.path.join(loc_vid_path, video_file) if video_file else "" | |
meta = annotations_dict.get(location_code, { | |
"spatial_type": spatial_type, | |
"index": -1, | |
"shop_names": "", | |
"sign_text": "", | |
}) | |
img_meta = image_meta_dict.get(file_name, {}) | |
vid_meta = video_meta_dict.get(video_file, {}) if video_file else {} | |
yield id_, { | |
"image_path": os.path.join(loc_img_path, file_name), | |
"video_path": video_path, | |
"location_code": location_code, | |
"spatial_type": meta["spatial_type"], | |
"index": meta["index"], | |
"shop_names": meta["shop_names"], | |
"sign_text": meta["sign_text"], | |
"image_metadata": str(img_meta), | |
"video_metadata": str(vid_meta), | |
} | |
id_ += 1 | |