ISIA Food-500: A Dataset for Large-Scale Food Recognition via Stacked Global-Local Attention Network

Data collection

ISIA Food-500 consists of 399,726 food items.Each item includes the food name,food images. There are totally 500 kinds of food dishes.

The dataset and code can be downloaded from the following link:

http://123.57.42.89/Dataset_ict/ISIA_Food500_Dir/

Code Implementation

This is a PyTorch implementation of the ACMMM2020 paper "ISIA Food-500: A Dataset for Large-Scale Food Recognition via Stacked Global-Local Attention Network" (Weiqing Min, Linhu Liu, Zhiling Wang,Zhengdong Luo, Xiaoming Wei, Xiaolin Wei and Shuqiang Jiang).

Requirements

• python 2.7
• pytorch 0.4+

Datasets

1.Download ten compressed packages(from ISIA_Food500.z01 to ISIA_Food500.z10) to obtain the complete dataset. You can also download metadata_ISIAFood_500.zip to obtain the statistical data of the dataset.
2.The category list is:

name_of_image.jpg label_num\n		

e.g: Duck_soup_noodles/Duck_soup_noodles_0045.jpg 197 
		

Train the model

If you want to train the model, just run 'python train_model.py'. You may need to change the configurations in train_model.py. The parameter 'DIR_TRAIN_IMAGES' is the category list of train and 'DIR_TEST_IMAGES' is the category list of test. 'WEIGHT_PATH' is the pretrained model on food dataset. 'IMAGE_PATH' is the path of the image folder. During training, the checkpoint file will be saved.

Reference

If you are interested in our work and want to cite it, please acknowledge the following paper:

@InProceedings{Min-ISIA-500-MM2020, author = {Weiqing Min and Linhu Liu and Zhiling Wang and Zhengdong Luo and Xiaoming Wei and Xiaolin Wei and Shuqiang Jiang}, title = {ISIA Food-500: A Dataset for Large-Scale Food Recognition via Stacked Global-Local Attention Network}, booktitle = {Proceedings of the 28th ACM International Conference on Multimedia}, year = {2020}, }