search

wuyuebupt / LFWBenchmark

5 stars 3 forks source link

The model of 99.53% on lfw #1

Open wuqiangch opened 7 years ago

wuqiangch commented 7 years ago

@wuyuebupt Can you share the model of 99.53% on lfw? And what's the database you used?

wuyuebupt commented 7 years ago

@wuqiangch Will add the model soon. The database is MS-Celeb-1M (MS-Celeb-1M: A Dataset and Benchmark for Large-Scale Face Recognition, ECCV 16).

wuqiangch commented 7 years ago

@wuyuebupt Thanks!

wuyuebupt commented 7 years ago

@wuqiangch The model is at https://drive.google.com/file/d/0B0W0rPKOeIBydzgzRTZ1aktrN1k/view?usp=sharing .

wuqiangch commented 7 years ago

@wuyuebupt I can download it now Thanks！

wuqiangch commented 7 years ago

@wuyuebupt There are more than 1,000 identities' overlap between MS-Celeb-1M and LFW.Have you remove the identities' of overlap to train the model and test on lfw？

wuqiangch commented 7 years ago

@wuyuebupt What't the method you used to detect face landmarks?

wuyuebupt commented 7 years ago

@wuqiangch Nope, I directly used the data. About alignment, for testing, I use MTCNNv2(https://github.com/kpzhang93/MTCNN_face_detection_alignment). For training, no alignment was used since the faces are aligned by Microsoft. Using the same alignment method with testing will get higher accuracy but 99.5% is enough.

wuqiangch commented 7 years ago

OK，thanks.

发自网易邮箱大师 On 03/01/2017 22:18, WU Yue wrote:

@wuqiangch Nope, I directly used the data. About alignment, for testing, I use MTCNNv2(https://github.com/kpzhang93/MTCNN_face_detection_alignment). For training, no alignment was used since the faces are aligned by Microsoft. Using the same alignment method with testing will get higher accuracy but 99.5% is enough.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.