The ImageNet dataset and competition

Description

ImageNet is a computer vision dataset with 14,197,122 images, 21,841 Synsets indexed. Synset is a node in WordNet hierarchy that in turn is a set of synonyms. There is a competition held every year with 1,000 classes from this dataset. It has been the standard benchmark for assessing the performance of image classification algorithms.

In 2013, a deep learning based computer vision model got the number one spot. From then, only deep learning models have won the competition. The following is the top five error rate over the years in the competition:

You can notice that the accuracy has been increasing over the years, as well as the depth of the layers. Next, we will understand the models that are present in this graph.

Efficient Graph-Based Image Segmentation

The Microsoft ResNet-50 model

The Microsoft ResNet-50 model