Summary of Look Through Masks: Towards Masked Face Recognition with De-occlusion Distillation, by Chenyu Li and Shiming Ge and Daichi Zhang and Jia Li
Look Through Masks: Towards Masked Face Recognition with De-Occlusion Distillation
by Chenyu Li, Shiming Ge, Daichi Zhang, Jia Li
First submitted to arxiv on: 19 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper addresses the challenge of recognizing masked faces in real-world applications like video surveillance and urban governance. When using diverse masks, face recognition accuracy drops significantly due to incomplete appearance and ambiguous representation. The authors propose an end-to-end de-occlusion distillation framework for masked face recognition, inspired by recent progress on amodal perception. The framework consists of two modules: a generative adversarial network (GAN) that completes the face under the mask, eliminating appearance ambiguity, and a distillation module that transfers knowledge from a pre-trained general face recognition model to train a student model for completed faces using massive online synthesized face pairs. The authors represent teacher knowledge as structural relations among instances in multiple orders, serving as posterior regularization to enable adaptation. Experimental results on synthetic and realistic datasets demonstrate the effectiveness of this approach. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine trying to recognize someone’s face when they’re wearing a mask that makes their features blurry or hard to see. This can happen in real-life situations like video surveillance or urban governance. The problem is that our current methods for recognizing faces don’t work well with masks, because the appearance of the face changes and becomes ambiguous. To solve this issue, researchers have developed an innovative approach called de-occlusion distillation. It involves two steps: first, they use a special network to complete the face under the mask, making it look more clear; then, they transfer knowledge from a well-trained model for recognizing faces in general to train a new model specifically designed for completed masked faces. The results show that this approach is very effective at recognizing masked faces. |
Keywords
» Artificial intelligence » Distillation » Face recognition » Gan » Generative adversarial network » Mask » Regularization » Student model
