Image recognition is the ability of software to identify objects, places, people, writing and actions in the images. 

Researchers from the Shenzhen Institutes of Advanced Technology (SIAT) of the Chinese Academy of Sciences proposed a product image recognition with guidance learning and noisy supervision, which showed advanced performance of daily product image recognition. The study has published in Computer Vision and Image Understanding. 

Instead of collecting product images by laborious and time-intensive image capturing, the team introduced a novel large-scale dataset, termed as Product-90, related to Clothing1M (a public large-scale dataset designed for learning from noisy data with human supervision) but contained much more categories, which was collected from the reviews of e-commerce websites, and it consisted of more than 140K images with 90 categories. 

In order to avoid the unrelated images, researchers further developed a simple yet efficient guidance learning (GL) method for training convolutional neural networks (CNNs) with noisy supervision. 

The team conducted comprehensive evaluations with this proposed guidance learning method on the Products-90 and four public datasets, namely Food101, Food-101N, Clothing1M and synthetic noisy CIFAR-10. 

At the first stage, they trained a baseline CNN model (teacher model) on the full Product-90 dataset (without the clean test set). At the second stage, they trained a target network (student network) on the large-scale noisy set and the small clean training set with multi-task learning. 

The results exhibited that this proposed guidance learning method was more efficient and simpler, and has achieved performance superior to state-of-the-art methods on these datasets.  

Media Contact: 
ZHANG Xiaomin 
Email: xm.zhang@siat.ac.cn