Skip to content
Showcase/BDD100K: Diverse Driving Dataset for Autonomous Driving
advancedimage

BDD100K: Diverse Driving Dataset for Autonomous Driving

BDD100K is a 100,000-video driving dataset from UC Berkeley (Yu et al., CVPR 2020) spanning detection, segmentation, lane marking, and tracking. This Potato config reproduces its driving-scene labeling tasks.

About this dataset

BDD100K is a large-scale, diverse driving dataset released by Fisher Yu and colleagues at UC Berkeley and presented at CVPR 2020. It was built to support heterogeneous multitask learning for autonomous driving perception, covering many cities, weather conditions, and times of day.

The dataset contains 100,000 driving videos, each about 40 seconds long, recorded at 720p (1280x720) and 30 fps. One frame near the 10th second of each video yields 100,000 still images used for image-level labeling, and a separate set of 10,000 images carries pixel-level segmentation labels.

Annotations span ten tasks: image tagging, object detection, lane marking, drivable-area segmentation, semantic segmentation, instance segmentation, panoptic segmentation, multi-object tracking (MOT), multi-object segmentation tracking (MOTS), plus domain adaptation and imitation learning. Detection uses 10 object categories and segmentation covers 40 classes.

The Potato config below reproduces the driving-scene labeling workflow with multiselect and radio schemes, letting annotators tag objects, lane markings, drivable areas, and scene attributes on individual frames.

Driving videos
100,000 (~40s each)
Labeled images
100,000 frames
Segmentation images
10,000 (pixel-level)
Detection categories
10 object classes
Segmentation classes
40 classes
Resolution / rate
720p at 30 fps
Labels:outdoornatureurbanpeopleanimal+−

Configuration Fileconfig.yaml

This Potato config reproduces the annotation task. Save it as config.yaml and run potato start config.yaml to try it.

yaml
# BDD100K Autonomous Driving Segmentation Configuration
# Based on Yu et al., CVPR 2020

annotation_task_name: "BDD100K Driving Scene Annotation"
task_dir: "."

data_files:
  - "sample-data.json"

item_properties:
  id_key: "id"
  text_key: "image_url"
  context_key: "context"

user_config:
  allow_all_users: true

annotation_schemes:
  - annotation_type: "multiselect"
    name: "objects"
    description: "Select all object types visible"
    labels:
      - name: "pedestrian"
        tooltip: "People walking"
      - name: "rider"
        tooltip: "Person on bike/motorcycle"
      - name: "car"
        tooltip: "Cars"
      - name: "truck"
        tooltip: "Trucks"
      - name: "bus"
        tooltip: "Buses"
      - name: "train"
        tooltip: "Trains"
      - name: "motorcycle"
        tooltip: "Motorcycles"
      - name: "bicycle"
        tooltip: "Bicycles"
      - name: "traffic_light"
        tooltip: "Traffic lights"
      - name: "traffic_sign"
        tooltip: "Traffic signs"

  - annotation_type: "multiselect"
    name: "lane_markings"
    description: "Select lane marking types"
    labels:
      - name: "road_curb"
        tooltip: "Road curb/edge"
      - name: "double_white"
        tooltip: "Double white lines"
      - name: "double_yellow"
        tooltip: "Double yellow lines"
      - name: "single_white"
        tooltip: "Single white line"
      - name: "single_yellow"
        tooltip: "Single yellow line"
      - name: "crosswalk"
        tooltip: "Crosswalk markings"

  - annotation_type: "multiselect"
    name: "drivable_area"
    description: "Select drivable area types"
    labels:
      - name: "direct"
        tooltip: "Directly drivable (same direction)"
      - name: "alternative"
        tooltip: "Alternatively drivable (lane change possible)"

  - annotation_type: "radio"
    name: "weather"
    description: "Weather condition"
    labels:
      - name: "clear"
        tooltip: "Clear weather"
      - name: "partly_cloudy"
        tooltip: "Partly cloudy"
      - name: "overcast"
        tooltip: "Overcast"
      - name: "rainy"
        tooltip: "Rainy"
      - name: "snowy"
        tooltip: "Snowy"
      - name: "foggy"
        tooltip: "Foggy"

  - annotation_type: "radio"
    name: "scene"
    description: "Scene type"
    labels:
      - name: "city_street"
        tooltip: "City street"
      - name: "highway"
        tooltip: "Highway"
      - name: "residential"
        tooltip: "Residential area"
      - name: "parking_lot"
        tooltip: "Parking lot"
      - name: "tunnel"
        tooltip: "Tunnel"

  - annotation_type: "radio"
    name: "time_of_day"
    description: "Time of day"
    labels:
      - name: "daytime"
        tooltip: "Daytime"
      - name: "dawn_dusk"
        tooltip: "Dawn or dusk"
      - name: "night"
        tooltip: "Nighttime"

interface_config:
  item_display_format: "<img src='{{text}}' style='max-width:100%; max-height:500px;'/><br/><small>{{context}}</small>"

output_annotation_format: "json"
output_annotation_dir: "annotations"

Sample Datasample-data.json

json
[
  {
    "id": "bdd_001",
    "image_url": "https://upload.wikimedia.org/wikipedia/commons/thumb/5/5a/Spyker_C8_Spyder.jpg/1200px-Spyker_C8_Spyder.jpg",
    "context": "Driving scene from dash cam. Annotate objects, lane markings, drivable areas, weather, and time of day."
  },
  {
    "id": "bdd_002",
    "image_url": "https://upload.wikimedia.org/wikipedia/commons/thumb/9/9e/Frankfurt_Skyline.jpg/1200px-Frankfurt_Skyline.jpg",
    "context": "Urban driving scene. Mark vehicles, pedestrians, traffic signs, and lane markings."
  }
]

// ... and 1 more items

Get This Design

View on GitHub

Clone or download from the repository

Quick start:

git clone https://github.com/davidjurgens/potato-showcase.git
cd potato-showcase/image/driving/bdd100k
potato start config.yaml

Dataset & paper

Yu et al., CVPR 2020

Official dataset ↗Read the paper ↗

Citation (BibTeX)

bibtex
@inproceedings{yu2020bdd100k,
    title = "{BDD100K}: A Diverse Driving Dataset for Heterogeneous Multitask Learning",
    author = "Yu, Fisher and Chen, Haofeng and Wang, Xin and Xian, Wenqi and Chen, Yingying and Liu, Fangchen and Madhavan, Vashisht and Darrell, Trevor",
    booktitle = "Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition",
    pages = "2633--2642",
    year = "2020",
    url = "https://arxiv.org/abs/1805.04687"
}

Details

Annotation Types

multiselectradio

Domain

Computer VisionAutonomous Driving

Use Cases

Object DetectionLane DetectionDrivable Area Segmentation

Tags

bdd100kdrivingautonomous-vehiclesegmentationcvpr2020

Found an issue or want to improve this design?

Open an Issue

Related Designs

KITTI Road Object Detection

Autonomous driving benchmark for object detection (Geiger et al., CVPR 2012). Annotate vehicles, pedestrians, and cyclists with 3D bounding boxes and occlusion/truncation labels.

multiselectradio

ADE20K: Scene Parsing and Semantic Segmentation Dataset

ADE20K is a scene parsing dataset from MIT CSAIL (Zhou et al., CVPR 2017) with dense pixel labels for objects, stuff, and parts. This Potato config adapts it to an image-level category-tagging task over a curated subset of common ADE20K categories.

multiselectradio

CUB-200-2011: Fine-Grained Bird Species Dataset

CUB-200-2011 is a fine-grained bird image dataset of 11,788 photos across 200 species, each labeled with parts, attributes, and a bounding box (Wah et al., 2011). This Potato config reproduces the labeling task.

multiselectradio