Datasets of Dish Images from Geolocalized Restaurants

Ruihan Xu, Shuqiang Jiang

Institute of Computing Technology, CAS


Descriptions

The datasets consist of dish images from restaurants with location information, thus can serve algorithms for dish recognition with visual and geograpical information.
1) Images are hierarchical organized: Restaurant-(Menu,Location)-Dish Image (Fig. 1). For each restaurant, there is a corresponding geographical coodinate which can uniquely locate it, and a menu which includes at least 3 dish categories. For each dish category, more than 15 images are included.
2) All the images are collected in 6 cities from an online restaurant review site. Table I shows the overall statistic of the datasets.

Figure 1: Examples of restaurants and dishes photos in the dataset (Beijing), which is hierarchical organized: Restaurant- (Menu, Location) -Dish Image.

Table I: OVERALL STATISTICS OF THE DATASETS (6CITIES).

Resources

Datasets:

Dish Images (.ZIP): Beijing Shanghai Tianjin Nanjing Hangzhou Guangzhou

Geographical Coordinates (.CSV): Beijing Shanghai Tianjin Nanjing Hangzhou Guangzhou

Other Charicteristics

1) Locations of the restaurants in our datasets are distributed across the city (See Fig.2).

Figure 2: Geographic distribution of the restaurants.
The background is a map of Beijing.

2) The number of dishes per restaurant and images per dish varies significantly across the datasets (See Fig.3). There are at least 3 dishes on the menu of each restaurant, and at least 15 images are included in each dish category.


(a)


(b)

Figure 3: (a) number of dish categories per restaurant, (b) number of images per dish category.(Data in "Beijing")


Contact: ruihan.xu [at] vipl.ict.ac.cn

Key Laboratory of Intelligent Information Processing
Institute of Computing Technology, CAS