Partial-FC
Partial FC is a distributed deep learning training framework for face recognition. The goal of Partial FC is to facilitate large-scale classification task (e.g. 10 or 100 million identities). It is much faster than the model parallel solution and there is no performance drop.
Contents
Glint360K
We clean, merge, and release the largest and cleanest face recognition dataset Glint360K, which contains 18 million images of 360K individuals. By employing the Patial FC training strategy, baseline models trained on Glint360K can easily achieve state-of-the-art performance. Detailed evaluation results on the large-scale test set (e.g. IFRT, IJB-C and Megaface) are as follows:
Evaluation on IFRT
r denotes the sampling rate of negative class centers.
| Backbone | Dataset | African | Caucasian | Indian | Asian | ALL |
|---|---|---|---|---|---|---|
| R50 | MS1M-V3 | 76.24 | 86.21 | 84.44 | 37.43 | 71.02 |
| R124 | MS1M-V3 | 81.08 | 89.06 | 87.53 | 38.40 | 74.76 |
| R100 | Glint360k(r=1.0) | 89.50 | 94.23 | 93.54 | 65.07 | 88.67 |
| R100 | Glint360k(r=0.1) | 90.45 | 94.60 | 93.96 | 63.91 | 88.23 |
Evaluation on IJB-C and Megaface
We employ ResNet100 as the backbone and CosFace (m=0.4) as the loss function. TAR@FAR=1e-4 is reported on the IJB-C datasets, and TAR@FAR=1e-6 is reported on the Megaface dataset.
| Test Dataset | IJB-C | Megaface_Id | Megaface_Ver |
|---|---|---|---|
| MS1MV2 | 96.4 | 98.3 | 98.6 |
| Glint360k | 97.3 | 99.1 | 99.1 |
Download
Baidu Drive (code:i1al)
Refer to the following command to unzip.
8b469dabd959b51f5ae63f1113fa9a7e glint360k00
25449911e03766bf773b12abcc5789cd glint360k01
bbd236acb4b561def6071f94b423d85e glint360k02
bd392b1506dddc5cead1b9de1e7b6e52 glint360k03
163aec95c3e258a79df7f1cb563fd5ef glint360k04
41bbf8314e24f98405ecbad8e9c89dfd glint360k05
dcc2f021aa2a9463ff1b2a8021ef87b6 glint360k06
cat glint360k* > glint360k.tar
tar -xvf glint360k.tar
# md5sum:
# train.rec 2a74c71c4d20e770273f103eda97e878
# train.idx f7a3e98d3533ac481bdf3dc03a5416e8
Use unpack_glint360k.py to unpack.
Performance
We neglect the influence of IO. All experiments use mixed-precision training, and the backbone is ResNet50.
1 Million Identities On 8 RTX2080Ti
| Method | GPUs | BatchSize | Memory/M | Throughput img/sec | W |
|---|---|---|---|---|---|
| Model Parallel | 8 | 1024 | 10408 | 2390 | GPU |
| Partial FC(Ours) | 8 | 1024 | 8100 | 2780 | GPU |
10 Million Identities On 64 RTX2080Ti
| Method | GPUs | BatchSize | Memory/M | Throughput img/sec | W |
|---|---|---|---|---|---|
| Model Parallel | 64 | 2048 | 9684 | 4483 | GPU |
| Partial FC(Ours) | 64 | 4096 | 6722 | 12600 | GPU |
Citation
If you find Partial-FC or Glint360K useful in your research, please consider to cite the following related paper:
@inproceedings{an2020partical_fc,
title={Partial FC: Training 10 Million Identities on a Single Machine},
author={An, Xiang and Zhu, Xuhan and Xiao, Yang and Wu, Lan and Zhang, Ming and Gao, Yuan and Qin, Bin and
Zhang, Debing and Fu Ying},
booktitle={Arxiv 2010.05222},
year={2020}
}
