 Depression is a common mental illness that can cause significant harm to those affected. Recent research suggests that objective physical signs such as facial expressions can be used to detect depression. To address this issue, we developed a weekly supervised learning approach based on multiple instance learning, MIL. This approach was tested on 150 videos of 75 depressed and 75 healthy subjects. The results show that our method achieved 74.7% accuracy and 74.5% recall, outperforming the baseline by 10.1% and surpassing the best performance of MIL based methods by 2.1%. This shows that our method has great potential for detecting depression through facial expressions. This article was authored by Zishu and Shangguan, Zhenyu Lu, Gang Li, and others.