ePillID Dataset: A Low-Shot Fine-Grained Benchmark for Pill Identification

TL;DR

The paper introduces ePillID, a large public benchmark dataset for pill image recognition designed for low-shot, fine-grained classification, highlighting its challenges and baseline model performances.

Contribution

It provides the largest public dataset for pill recognition with a challenging low-shot setting and evaluates baseline models on this new benchmark.

Findings

Multi-head metric-learning approach performs well

Baseline models struggle with confusing classes

Dataset enables future research in pill identification

Abstract

Identifying prescription medications is a frequent task for patients and medical professionals; however, this is an error-prone task as many pills have similar appearances (e.g. white round pills), which increases the risk of medication errors. In this paper, we introduce ePillID, the largest public benchmark on pill image recognition, composed of 13k images representing 9804 appearance classes (two sides for 4902 pill types). For most of the appearance classes, there exists only one reference image, making it a challenging low-shot recognition setting. We present our experimental setup and evaluation results of various baseline models on the benchmark. The best baseline using a multi-head metric-learning approach with bilinear features performed remarkably well; however, our error analysis suggests that they still fail to distinguish particularly confusing classes. The code and data are available at https://github.com/usuyama/ePillID-benchmark.