Implemented the Deep Residual Learning for Image Recognition Paper and achieved better accuracy by customizing different parts of the architecture.

WarmUp(学习率预热)、CosineAnnealingWarmRestarts(带重启的余弦退火模型)

A deep learning project for 7-class Facial Expression Recognition with post-hoc explainability using LIME and SHAP. Two CNN architectures — ResNet-50 and EfficientNet-B0 — are trained and evaluated on FER2013 and RAF-DB, with pixel-level and region-level explanations generated for every prediction.