Skip to content
intermediateimage

Places365 Scene Classification

Scene recognition and classification following the Places365 dataset (Zhou et al., TPAMI 2017). Classify images into 365 scene categories spanning indoor, outdoor, and natural environments.

About this dataset

Scene recognition is the task of labeling a whole image with the kind of place or environment it depicts, such as a bedroom, a forest, or an airport terminal. The Places database was introduced by Bolei Zhou, Agata Lapedriza, Aditya Khosla, Aude Oliva, and Antonio Torralba in "Places: A 10 Million Image Database for Scene Recognition" (IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017). It was built to give scene-centric neural networks training data at the scale that object-centric datasets like ImageNet already had, so that models could learn to recognize environments rather than isolated objects.

The images were collected from web search engines using a large vocabulary of scene categories drawn from the SUN database, then passed to crowd workers who confirmed whether each image matched its category. The result is organized into several releases; Places365-Standard is the subset most commonly used for training and benchmarking, covering 365 scene categories that span indoor, urban, and natural environments.

Places365-Standard contains about 1.8 million training images across its 365 categories, with 50 validation images and 900 test images per category (18,250 validation and 328,500 test images in total). The full Places database holds over 10 million images. Category labels were verified through crowdsourced agreement to keep the scene assignments consistent.

The Potato config below reproduces a coarse-grained version of this task rather than the full 365-way vocabulary. It uses a radio scheme for a top-level scene category (eight groups such as indoor_home, outdoor_nature, and transportation), a multiselect for scene attributes (natural, man-made, open, enclosed, vegetation, water, crowded, and others), a second radio for indoor versus outdoor versus semi-outdoor, and a free-text field for a specific place name. This suits quick scene labeling and attribute collection when a full fine-grained taxonomy is not needed.

Scene categories (Places365)
365
Training images (Places365-Standard)
~1.8 million
Validation images
18,250 (50 per category)
Test images
328,500 (900 per category)
Full Places database
10+ million images
Venue / Year
IEEE TPAMI 2017 (vol. 40, no. 6, pp. 1452-1464)
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
# Places365 Scene Classification Configuration
# Based on Zhou et al., TPAMI 2017

annotation_task_name: "Places365 Scene Classification"
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: "radio"
    name: "scene_category"
    description: "Select the primary scene category"
    labels:
      - name: "indoor_home"
        tooltip: "Indoor home environments (bedroom, kitchen, living room)"
      - name: "indoor_work"
        tooltip: "Indoor work environments (office, conference room)"
      - name: "indoor_public"
        tooltip: "Indoor public spaces (mall, museum, restaurant)"
      - name: "outdoor_urban"
        tooltip: "Outdoor urban scenes (street, plaza, parking lot)"
      - name: "outdoor_nature"
        tooltip: "Natural outdoor scenes (forest, beach, mountain)"
      - name: "outdoor_sports"
        tooltip: "Sports venues (stadium, tennis court, golf course)"
      - name: "transportation"
        tooltip: "Transportation scenes (airport, train station, highway)"
      - name: "water"
        tooltip: "Water-related scenes (ocean, lake, river)"

  - annotation_type: "multiselect"
    name: "scene_attributes"
    description: "Select scene attributes that apply"
    labels:
      - name: "natural"
        tooltip: "Scene contains natural elements"
      - name: "man-made"
        tooltip: "Scene contains man-made structures"
      - name: "open"
        tooltip: "Open, expansive space"
      - name: "enclosed"
        tooltip: "Enclosed or bounded space"
      - name: "rugged"
        tooltip: "Uneven, rough terrain"
      - name: "smooth"
        tooltip: "Smooth, even surfaces"
      - name: "vegetation"
        tooltip: "Contains plants or greenery"
      - name: "water"
        tooltip: "Contains water"
      - name: "crowded"
        tooltip: "Contains many people"
      - name: "empty"
        tooltip: "Few or no people"

  - annotation_type: "radio"
    name: "indoor_outdoor"
    description: "Is this an indoor or outdoor scene?"
    labels:
      - name: "indoor"
        tooltip: "Scene is indoors"
      - name: "outdoor"
        tooltip: "Scene is outdoors"
      - name: "semi_outdoor"
        tooltip: "Partially covered or transitional space"

  - annotation_type: "text"
    name: "specific_place"
    description: "Enter a more specific place description if known"

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": "places_001",
    "image_url": "https://upload.wikimedia.org/wikipedia/commons/thumb/1/10/Empire_State_Building_%28aerial_view%29.jpg/800px-Empire_State_Building_%28aerial_view%29.jpg",
    "context": "Classify this scene. Consider whether it's indoor/outdoor and what category it belongs to."
  },
  {
    "id": "places_002",
    "image_url": "https://upload.wikimedia.org/wikipedia/commons/thumb/e/ea/Van_Gogh_-_Starry_Night_-_Google_Art_Project.jpg/1200px-Van_Gogh_-_Starry_Night_-_Google_Art_Project.jpg",
    "context": "Identify the scene type and applicable attributes."
  }
]

// ... 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/classification/places365
potato start config.yaml

Dataset & paper

Zhou et al., IEEE TPAMI 2017

Official dataset ↗Read the paper ↗

Citation (BibTeX)

bibtex
@article{zhou2017places,
    title = {Places: A 10 Million Image Database for Scene Recognition},
    author = {Zhou, Bolei and Lapedriza, Agata and Khosla, Aditya and Oliva, Aude and Torralba, Antonio},
    journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
    volume = {40},
    number = {6},
    pages = {1452--1464},
    year = {2017},
    publisher = {IEEE}
}

Details

Annotation Types

multiselectradiotext

Domain

Computer VisionScene Recognition

Use Cases

Scene ClassificationPlace RecognitionEnvironment Understanding

Tags

placessceneclassificationindoor-outdoortpami2017

Found an issue or want to improve this design?

Open an Issue

Related Designs

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

FLAIR: French Land Cover from Aerospace Imagery

Land use and land cover classification from high-resolution aerial imagery. Annotators classify the primary land use category of aerial image patches and identify any secondary land uses present. Based on the FLAIR dataset from the French National Institute of Geographic and Forest Information (IGN).

multiselectradio

iWildCam Wildlife Detection & Classification

Camera trap image classification for wildlife monitoring (Beery et al., CVPR 2019). Classify wildlife species from camera trap images across diverse ecosystems worldwide.

multiselectradio