Table of contents
1. Classify Network Datasets
1.1 ImageNet2012
1000 classifications, used for training and verification of the classification network (classification).
1.1.1 Training set
According to the category, it is divided into 1000 synsets, each synset corresponds to a category of pictures, and is named by the WNID of the synset. There is a corresponding relationship between synset WNID (wordNet ID) and ILSVRC2012_ID (1-1000), see ILSVRC2012_devkit_t12/data/meta.mat
- Total number of images: 1281167
- Memory size: 137.74GB
- Number of Synsets: ranging from 732-1300
- Image file naming format: x_y.JPEG
x: synset WNID;
y: non-fixed length non-sequential integer - ILSVRC2012_devkit_t12/data/meta.mat
contains: ILSCRC2012_ID, WNID, words, gloss, number_train_images
1.1.2. Validation set
There are 50,000 pictures, which are not classified according to category. The command form is as follows:
- Validation set label: ILSVRC2012_devkit_t12/data/ ILSVRC2012_validation_ground_truth.txt
- Label order corresponds to images_name
- Classes_id[1,1000]
- 50 samples per category
- Class description: ILSVRC2012_devkit_t12/data/meta.mat
- Memory size: 6.28GB