LaCoVL-FER: Landmark-Guided Contrastive Learning Network with Vision-Language Enhancement for Facial Expression Recognition

Researchers introduced LaCoVL-FER, a facial expression recognition model that combines facial landmarks with CLIP-based vision-language features.

The method uses landmark-guided adaptive encoding and vision-language enhancement to fuse geometric and semantic cues.

On RAF-DB, FERPlus, and AffectNet, the model delivered state-of-the-art performance.

The approach is designed to improve robustness under pose changes, occlusion, and lighting variation.