Fast and Accurate Quantized Camera Scene Detection on Smartphones, Mobile AI 2021 Challenge: Report

Camera scene detection is among the most popular computer vision problem on smartphones. While many custom solutions were developed for this task by phone vendors, none of the designed models were available publicly up until now. To address this problem, we introduce the first Mobile AI challenge, where the target is to develop quantized deep learning-based camera scene classification solutions that can demonstrate a real-time performance on smartphones and IoT platforms. For this, the participants were provided with a large-scale CamSDD dataset consisting of more than 11K images belonging to the 30 most important scene categories. The runtime of all models was evaluated on the popular Apple Bionic A11 platform that can be found in many iOS devices. The proposed solutions are fully compatible with all major mobile AI accelerators and can demonstrate more than 100-200 FPS on the majority of recent smartphone platforms while achieving a top-3 accuracy of more than 98%. A detailed description of all models developed in the challenge is provided in this paper.

1. Introduction

The problem of automatic camera scene prediction on smartphones appeared soon after the introduction of the first mobile cameras. While the initial scene classification approaches were using only manually designed features and some simple machine learning algorithms, the availability of much more powerful AI hardware such as NPUs, GPUs and DSPs made it possible to use considerably more accurate and efficient deep learning-based solutions. Neverthe less, this task has not been properly addressed in the literature until the introduction of the Camera Scene Detection Dataset (CamSDD) dataset in [36], where this problem was carefully defined and training data for 30 different camera scene categories was provided along with a fast baseline solution. In this challenge, we take one step further in solving this task by imposing additional efficiency-related constraints on the developed models.