Romanian Traffic Webcams - 12K

Romanian Traffic Webcams-12K (RTW-12K) is a dataset of traffic-camera images captured from publicly accessible webcams across Romania, positioned slightly above street level or mounted on taller buildings. Data were collected via the Webcam Romania (2026) platform, which provides access to a wide range of urban and traffic webcams streamed in m3u8 format or via YouTube Live. We selected 12 locations and captured frames at fixed intervals (approximately every 15 seconds, depending on the camera), yielding roughly 20,000 images per location. These locations were chosen based on traffic density and camera elevation, with some sites specifically selected to increase representation for particular vehicle categories,such as border checkpoints for trucks or railway stations where busses frequently appear. The full collection was first processed using the YOLOv12x model to obtain preliminary annotations. Based on these predictions, we selected approximately 1,000 images per target class, acknowledging that each image may contain multiple classes to achieve a more balanced dataset. All selected images were then reviewed and manually corrected to ensure annotation accuracy.To further increase dataset diversity, particularly with respect to adverse weather, we repeated the acquisition process at an additional 12 locations (some overlapping with the initial set) during heavy rainfall. From this second round, approximately 200 images per class were manually selected.The final dataset contains 12,196 images.-------------------------------------------------------------------------------------------------------------## Dataset StructureThe dataset is organized as follows:dataset/│├── rain/│   ├── images_nodata/│   ├── labels/│   └── classes.txt│├── norain/│   ├── images_nodata/│   ├── labels/│   └── classes.txt### Folder Description*images_nodata/*Contains a readme explaining how one can get the image files used in the dataset.*labels/*Contains the annotation files corresponding to the images. Each label file has the same name as its corresponding image file.classes.txtLists all object classes used in the annotations.---## Annotation FormatAnnotations follow the YOLO object detection format.Each annotation file corresponds to one image and contains one object per line in the following format:<class_id> <x_center> <y_center> <width> <height>Where:* class_id – numeric identifier of the object class x_center – normalized x coordinate of the bounding box center* y_center – normalized y coordinate of the bounding box center width – normalized width of the bounding box* height – normalized height of the bounding boxAll coordinates are normalized between 0 and 1 relative to the image size.---## Object ClassesThe dataset includes the following object categories:0  awning-tricycle1  bicycle2  bus3  car4  motor5  pedestrian6  people7  scooter8  semi_trailer9  tricycle10 truck11 vanDescriptions: awning-tricycle – three-wheeled vehicle with canopy* bicycle – standard two-wheel bicycle bus – public transportation bus* car – passenger vehicle* motor – motorcycle or motorbike* pedestrian – single walking person* people – group of people* scooter – small two-wheel motor scooter* semi_trailer – articulated truck trailer* tricycle – three-wheel vehicle* truck – cargo truck* van – light commercial van---## Weather ConditionsThe dataset contains two weather categories:Rain* Images captured during rainy conditions.No Rain* Images captured in normal weather conditions without precipitation.This split allows researchers to study the impact of weather on object detection performance.---## Intended UseThe dataset can be used for:* Object detection training and evaluation* Weather-aware perception systems* Traffic monitoring research* Robust detection under adverse weather conditions---## Data AccessThe annotations are publicly available.The corresponding images may be available upon reasonable request for research purposes.Requests should include:* Full name* Institutional affiliation* Intended use of the dataset* Confirmation that the images will not be redistributedContact:[email protected]---## LicenseThe annotation files are released under the CC BY 4.0 license unless stated otherwise.The license applies to annotations and metadata only, not to the underlying images.---## ContactFor questions, data access requests, or collaborations:[email protected]