Pytorch implements the Deep Face Recognition part of Insightface(github) with a backbone of EfficientNet(github).
Official explanation: EfficientNets are a family of image classification models, which achieve state-of-the-art accuracy, yet being an order-of-magnitude smaller and faster than previous models. We develop EfficientNets based on AutoML and Compound Scaling. In particular, we first use AutoML Mobile framework to develop a mobile-size baseline network, named as EfficientNet-B0; Then, we use the compound scaling method to scale up this baseline to obtain EfficientNet-B1 to B7.
|
|
Details about the EfficientNet models are below:
| Name | # Params | Top-1 Acc. |
|---|---|---|
efficientnet-b0 |
5.3M | 76.3 |
efficientnet-b1 |
7.8M | 78.8 |
efficientnet-b2 |
9.2M | 79.8 |
efficientnet-b3 |
12M | 81.1 |
efficientnet-b4 |
19M | 82.6 |
efficientnet-b5 |
30M | 83.3 |
efficientnet-b6 |
43M | 84.0 |
efficientnet-b7 |
66M | 84.4 |
downloading the Training data MS1M, face is detected by MTCNN and resized to 112x112. If you need to tansfer the .bin or .rec files into images(.jpg),please run the script python GetImages.py under your data fold, note that maxnet should be install.
a. EfficientNet(b0,Params is 5.3M) with batchsize 80 + Argface(m=64,s=0.5) + focalloss(gam=2)
| LFW(%) | CFP-FF(%) | CFP-FP(%) | AgeDB-30(%) | calfw(%) | cplfw(%) | vgg2_fp(%) |
|---|---|---|---|---|---|---|
| 0.9955 | 0.9940 | 0.9347 | 0.9545 | 0.9532 | 0.8973 | 0.9320 |
b. EfficientNet(b7,Params is 66M) with batchsize 80 + Argface(m=64,s=0.5) + focalloss(gam=2)
The results is only trained 20 epoch, pretrained model can be download in here(baidu drive, code:wkd2) or here(google drive).
| LFW(%) | CFP-FF(%) | CFP-FP(%) | AgeDB-30(%) | calfw(%) | cplfw(%) | vgg2_fp(%) |
|---|---|---|---|---|---|---|
| 0.9973 | 0.9967 | 0.9620 | 0.9705 | 0.9553 | 0.9105 | 0.9428 |
c.other pretrained model b1, b2, ..., b6 and results is updating...
If you have questions, post them as GitHub issues.