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.
About this dataset
ADE20K is a scene parsing and semantic segmentation dataset built by Bolei Zhou and colleagues at MIT CSAIL and presented at CVPR 2017. Every image is densely labeled at the pixel level, covering objects, background stuff, and the parts that make up each object.
The full dataset contains 25,210 images: 20,210 for training, 2,000 for validation, and 3,000 for testing. Across these images annotators marked 434,826 object instances and 175,961 part instances, with an average of 19.5 instances and 10.5 object classes per image.
The annotations span 2,693 object and stuff classes plus 476 part classes, with parts attached to a large share of object instances. The widely used SceneParse150 benchmark narrows this to 150 frequent classes (115 objects and 35 stuff categories), which cover 92.75% of all annotated pixels and serve as the standard semantic segmentation evaluation set.
The Potato config below reproduces this labeling task. Annotators select which of the 150 scene categories appear in an image and assign a primary scene label, mirroring the object-plus-stuff taxonomy that drives ADE20K's pixel-level segmentation.
- Total images
- 25,210
- Train / val split
- 20,210 / 2,000
- Object + stuff classes
- 2,693
- Part classes
- 476
- SceneParse150 benchmark
- 150 classes
- Object instances
- 434,826
Configuration Fileconfig.yaml
This Potato config reproduces the annotation task. Save it as config.yaml and run potato start config.yaml to try it.
# ADE20K Semantic Segmentation Configuration
# Based on Zhou et al., CVPR 2017
annotation_task_name: "ADE20K Scene Parsing"
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: "building_elements"
description: "Select visible building/structure elements"
labels:
- name: "wall"
tooltip: "Walls"
- name: "floor"
tooltip: "Floor surfaces"
- name: "ceiling"
tooltip: "Ceilings"
- name: "door"
tooltip: "Doors"
- name: "window"
tooltip: "Windows"
- name: "stairs"
tooltip: "Stairs, steps"
- annotation_type: "multiselect"
name: "furniture"
description: "Select visible furniture"
labels:
- name: "chair"
tooltip: "Chairs, seats"
- name: "table"
tooltip: "Tables"
- name: "bed"
tooltip: "Beds"
- name: "sofa"
tooltip: "Sofas, couches"
- name: "cabinet"
tooltip: "Cabinets, shelves"
- name: "desk"
tooltip: "Desks"
- annotation_type: "multiselect"
name: "nature"
description: "Select visible nature elements"
labels:
- name: "tree"
tooltip: "Trees"
- name: "grass"
tooltip: "Grass"
- name: "plant"
tooltip: "Plants, flowers"
- name: "sky"
tooltip: "Sky"
- name: "water"
tooltip: "Water bodies"
- name: "mountain"
tooltip: "Mountains, hills"
- annotation_type: "multiselect"
name: "objects"
description: "Select visible objects"
labels:
- name: "lamp"
tooltip: "Lamps, lights"
- name: "painting"
tooltip: "Paintings, artwork"
- name: "curtain"
tooltip: "Curtains, drapes"
- name: "rug"
tooltip: "Rugs, carpets"
- name: "mirror"
tooltip: "Mirrors"
- name: "clock"
tooltip: "Clocks"
- annotation_type: "radio"
name: "scene_type"
description: "What type of scene is this?"
labels:
- name: "indoor"
tooltip: "Indoor scene"
- name: "outdoor"
tooltip: "Outdoor scene"
- name: "mixed"
tooltip: "Mixed indoor/outdoor"
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
[
{
"id": "ade_001",
"image_url": "https://upload.wikimedia.org/wikipedia/commons/thumb/5/5e/Schloss_Neuschwanstein_2013.jpg/1200px-Schloss_Neuschwanstein_2013.jpg",
"context": "Parse this scene by identifying all visible objects, structures, and nature elements."
},
{
"id": "ade_002",
"image_url": "https://upload.wikimedia.org/wikipedia/commons/thumb/a/a7/Camponotus_flavomarginatus_ant.jpg/1200px-Camponotus_flavomarginatus_ant.jpg",
"context": "Annotate all semantic categories visible in this scene."
}
]
// ... and 1 more itemsGet This Design
Clone or download from the repository
Quick start:
git clone https://github.com/davidjurgens/potato-showcase.git cd potato-showcase/image/segmentation/ade20k potato start config.yaml
Dataset & paper
Zhou et al., CVPR 2017
Citation (BibTeX)
@inproceedings{zhou2017scene,
title = "Scene Parsing through {ADE20K} Dataset",
author = "Zhou, Bolei and Zhao, Hang and Puig, Xavier and Fidler, Sanja and Barriuso, Adela and Torralba, Antonio",
booktitle = "Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)",
pages = "633--641",
year = "2017",
url = "https://openaccess.thecvf.com/content_cvpr_2017/html/Zhou_Scene_Parsing_Through_CVPR_2017_paper.html"
}Details
Annotation Types
Domain
Use Cases
Tags
Found an issue or want to improve this design?
Open an IssueRelated Designs
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.
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.
DeepFashion: Large-Scale Clothing Attributes Dataset
DeepFashion is a clothes recognition and retrieval dataset of 800,000+ images labeled with 50 categories, 1,000 attributes, and landmarks (Liu et al., CVPR 2016). This Potato config reproduces its category and attribute task.