ViT-FREE improves face recognition efficiency via early exiting

Researchers have developed ViT-FREE, a method to make Vision Transformers more efficient for face recognition. This approach allows for early exiting from intermediate layers of a pre-trained ViT, reducing computational cost without retraining the model. Experiments show that exiting at later layers provides a good balance of speed and accuracy, with up to a 20% speedup on benchmarks like IJB-C at a minimal performance drop. An additional fine-tuning strategy, ViT-FREE_FT, further optimizes shallow exits by adapting only projection layers with synthetic data. AI

IMPACT Enhances efficiency for Vision Transformer-based face recognition models, potentially enabling deployment on resource-constrained devices.

RANK_REASON The cluster contains a research paper detailing a new method for improving model efficiency. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• IJB-C
• Vision Transformers
• ViT-FREE

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →