semantic-and-visual-similarities-for-efficient-knowledge-transfer-in-CNN-training

This repository the constitution of the dataset used in ...

• synsets.txt : The 90 Imagenet synsets of the dataset (out of the ILSVRC challenge).
• train.txt : The 81000 images selected for the training set (900 per synset).
• test.txt : The 9000 images selected for the test set (100 per synset).

For more details about the synsets' and images' selection process, please refer to the paper.