paper
loss
face recognition
轻量级网络
量化
剪枝
- NISP: Pruning Networks using Neuron Importance Score Propagation
- Net-Trim: Convex Pruning of Deep Neural Networks with Performance Guarantee
- Channel Pruning for Accelerating Very Deep Neural Networks
vertification
LFW和MegaFace均为人脸识别领域重要的评测数据集。LFW是针对早期人脸验证(Face Verification)任务提出评测方法与指标,结果有借鉴意义,但已不代表目前的最难问题。 MegaFace提出的关于百万级别的1:N人脸辨识(Face Identification)任务的评测指标,难度更大,是目前学术界的新主流。