Mastering Large Scale Multi-label Image Recognition with high efficiency overCamera trap images

TL;DR

This paper presents a fast, lightweight machine learning approach for large-scale multi-label image recognition in camera trap images, achieving high accuracy and outperforming humans with minimal data augmentation.

Contribution

The authors introduce an efficient, simple baseline method that handles large datasets with limited hardware, avoiding overfitting and achieving top performance.

Findings

Achieved 97% accuracy on wildlife identification

Outperformed human-level performance

Trained on 6.7 million images using a single GPU

Abstract

Camera traps are crucial in biodiversity motivated studies, however dealing with large number of images while annotating these data sets is a tedious and time consuming task. To speed up this process, Machine Learning approaches are a reasonable asset. In this article we are proposing an easy, accessible, light-weight, fast and efficient approach based on our winning submission to the "Hakuna Ma-data - Serengeti Wildlife Identification challenge". Our system achieved an Accuracy of 97% and outperformed the human level performance. We show that, given relatively large data sets, it is effective to look at each image only once with little or no augmentation. By utilizing such a simple, yet effective baseline we were able to avoid over-fitting without extensive regularization techniques and to train a top scoring system on a very limited hardware featuring single GPU (1080Ti) despite the large training set (6.7M images and 6TB).