Skip to content

GitLab

Commit 8760acf2 authored by dongchy920's avatar dongchy920
Browse files

arcface_pytorch

parent 9b87ce51
Hide whitespace changes
Inline Side-by-side
Showing with 182 additions and 338 deletions
LICENSE.md 0 → 100644
View file @ 8760acf2
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction, and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all other entities that control, are controlled by, or are under common control with that entity. For the purposes of this definition, "control" means (i) the power, direct or indirect, to cause the direction or management of such entity, whether by contract or otherwise, or (ii) ownership of fifty percent (50%) or more of the outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications, including but not limited to software source code, documentation source, and configuration files.
"Object" form shall mean any form resulting from mechanical transformation or translation of a Source form, including but not limited to compiled object code, generated documentation, and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or Object form, made available under the License, as indicated by a copyright notice that is included in or attached to the work (an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object form, that is based on (or derived from) the Work and for which the editorial revisions, annotations, elaborations, or other modifications represent, as a whole, an original work of authorship. For the purposes of this License, Derivative Works shall not include works that remain separable from, or merely link (or bind by name) to the interfaces of, the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including the original version of the Work and any modifications or additions to that Work or Derivative Works thereof, that is intentionally submitted to Licensor for inclusion in the Work by the copyright owner or by an individual or Legal Entity authorized to submit on behalf of the copyright owner. For the purposes of this definition, "submitted" means any form of electronic, verbal, or written communication sent to the Licensor or its representatives, including but not limited to communication on electronic mailing lists, source code control systems, and issue tracking systems that are managed by, or on behalf of, the Licensor for the purpose of discussing and improving the Work, but excluding communication that is conspicuously marked or otherwise designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity on behalf of whom a Contribution has been received by Licensor and subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable (except as stated in this section) patent license to make, have made, use, offer to sell, sell, import, and otherwise transfer the Work, where such license applies only to those patent claims licensable by such Contributor that are necessarily infringed by their Contribution(s) alone or by combination of their Contribution(s) with the Work to which such Contribution(s) was submitted. If You institute patent litigation against any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Work or a Contribution incorporated within the Work constitutes direct or contributory patent infringement, then any patent licenses granted to You under this License for that Work shall terminate as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the Work or Derivative Works thereof in any medium, with or without modifications, and in Source or Object form, provided that You meet the following conditions:
You must give any other recipients of the Work or Derivative Works a copy of this License; and
You must cause any modified files to carry prominent notices stating that You changed the files; and
You must retain, in the Source form of any Derivative Works that You distribute, all copyright, patent, trademark, and attribution notices from the Source form of the Work, excluding those notices that do not pertain to any part of the Derivative Works; and
If the Work includes a "NOTICE" text file as part of its distribution, then any Derivative Works that You distribute must include a readable copy of the attribution notices contained within such NOTICE file, excluding those notices that do not pertain to any part of the Derivative Works, in at least one of the following places: within a NOTICE text file distributed as part of the Derivative Works; within the Source form or documentation, if provided along with the Derivative Works; or, within a display generated by the Derivative Works, if and wherever such third-party notices normally appear. The contents of the NOTICE file are for informational purposes only and do not modify the License. You may add Your own attribution notices within Derivative Works that You distribute, alongside or as an addendum to the NOTICE text from the Work, provided that such additional attribution notices cannot be construed as modifying the License.
You may add Your own copyright statement to Your modifications and may provide additional or different license terms and conditions for use, reproduction, or distribution of Your modifications, or for any such Derivative Works as a whole, provided Your use, reproduction, and distribution of the Work otherwise complies with the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise, any Contribution intentionally submitted for inclusion in the Work by You to the Licensor shall be under the terms and conditions of this License, without any additional terms or conditions. Notwithstanding the above, nothing herein shall supersede or modify the terms of any separate license agreement you may have executed with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade names, trademarks, service marks, or product names of the Licensor, except as required for reasonable and customary use in describing the origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or agreed to in writing, Licensor provides the Work (and each Contributor provides its Contributions) on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied, including, without limitation, any warranties or conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE. You are solely responsible for determining the appropriateness of using or redistributing the Work and assume any risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory, whether in tort (including negligence), contract, or otherwise, unless required by applicable law (such as deliberate and grossly negligent acts) or agreed to in writing, shall any Contributor be liable to You for damages, including any direct, indirect, special, incidental, or consequential damages of any character arising as a result of this License or out of the use or inability to use the Work (including but not limited to damages for loss of goodwill, work stoppage, computer failure or malfunction, or any and all other commercial damages or losses), even if such Contributor has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing the Work or Derivative Works thereof, You may choose to offer, and charge a fee for, acceptance of support, warranty, indemnity, or other liability obligations and/or rights consistent with this License. However, in accepting such obligations, You may act only on Your own behalf and on Your sole responsibility, not on behalf of any other Contributor, and only if You agree to indemnify, defend, and hold each Contributor harmless for any liability incurred by, or claims asserted against, such Contributor by reason of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
\ No newline at end of file
README.md
View file @ 8760acf2
# Distributed Arcface Training in Pytorch
The "arcface_torch" repository is the official implementation of the ArcFace algorithm. It supports distributed and sparse training with multiple distributed training examples, including several memory-saving techniques such as mixed precision training and gradient checkpointing. It also supports training for ViT models and datasets including WebFace42M and Glint360K, two of the largest open-source datasets. Additionally, the repository comes with a built-in tool for converting to ONNX format, making it easy to submit to MFR evaluation systems.
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/killing-two-birds-with-one-stone-efficient/face-verification-on-ijb-c)](https://paperswithcode.com/sota/face-verification-on-ijb-c?p=killing-two-birds-with-one-stone-efficient)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/killing-two-birds-with-one-stone-efficient/face-verification-on-ijb-b)](https://paperswithcode.com/sota/face-verification-on-ijb-b?p=killing-two-birds-with-one-stone-efficient)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/killing-two-birds-with-one-stone-efficient/face-verification-on-agedb-30)](https://paperswithcode.com/sota/face-verification-on-agedb-30?p=killing-two-birds-with-one-stone-efficient)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/killing-two-birds-with-one-stone-efficient/face-verification-on-cfp-fp)](https://paperswithcode.com/sota/face-verification-on-cfp-fp?p=killing-two-birds-with-one-stone-efficient)
## Requirements
To avail the latest features of PyTorch, we have upgraded to version 1.12.0.
- Install [PyTorch](https://pytorch.org/get-started/previous-versions/) (torch>=1.12.0).
- (Optional) Install [DALI](https://docs.nvidia.com/deeplearning/dali/user-guide/docs/), our doc for [install_dali.md](docs/install_dali.md).
- `pip install -r requirement.txt`.
## How to Training
To train a model, execute the `train_v2.py` script with the path to the configuration files. The sample commands provided below demonstrate the process of conducting distributed training.
### 1. To run on one GPU:
```shell
python train_v2.py configs/ms1mv3_r50_onegpu
# ArcFace
## 论文
- https://arxiv.org/pdf/1801.07698.pdf
## 模型结构
这篇文章提出一种新的用于人脸识别的损失函数：additive angular margin loss，直接在角度空间（angular space）中最大化分类界限，基于该损失函数训练得到人脸识别算法ArcFace。
<div align=center>
<img src="./docs/arcface.png"/>
</div>
## 算法原理
通过训练深度卷积神经网络嵌入 (DCNN Embedding) 来进行人脸识别。
ArcFace训练流程：
设类别数（人脸ID数量）为 $n$,DCNN的最后一个FC 层的权重为$W\subset {\mathbb{R}}^{d \times n}$，输入$W$的特征$x_i$的维度为$d$。
1、分别归一化输入特征$x_i \subset {\mathbb{R}}^{b}$和FC层权重$W_j \in {\mathbb{R}}^{1 \times b}$（张量除以欧几里得范数标量），令所得归一化特征$\frac{x_i}{\|x_i\|}$与第$j \in {1,2,...,y_i,...,n}$个类别的FC层权重$\frac{{W_j}^T}{\|W_j\|} \in {\mathbb{R}}^{1\times d}$点乘得到FC层的第$j$个输出$cos \theta_j \in {\mathbb{R}}^{1\times1}$（数量积公式：${W_j}^{T}\cdot x_i=\|W_j\|\|x_i\|cos\theta_j$）,表示**将特征$x_i$预测为第$j$类的预测值**。
2、设特征$x_i$的真实类别为第$y_j$个类别，单独取出Target权重$\frac{{W_{y_j}}^T}{\|W_{y_i}\|}$计算$\theta_{y_i}=arccos(cos\theta_{y_i})=arccos(\frac{{W_{y_j}}^T}{\|W_{y_i}\|}\cdot\frac{x_i}{\|x_i\|})$可得归一化特征$\frac{x_i}{\|x_i\|}$与归一化**target权重**$\frac{{W_{y_j}}^T}{\|W_{y_i}\|}$之间的夹角—— **Target角度$\theta_{y_i}$**
3、通过把一个自定义的**加性角度边距 (additive angular margin)** $m$加到$\theta_{y_i}$，得到$\theta_{y_i}+m$，用于**调整Target角度**。
4、计算经调整的Target角度的余弦，得到仅关于特征$x_i$的真实类别$y_i$的**新Target Logit $cos(\theta_{y_i}+m)$**。
5、通过自定义的特征范数$s$重缩放所有Logit（除Target Logit变为$cos(\theta_{y_i}+m)$）外其余原Logit仍为$cos\theta_j$，矩阵运算时需用相当于 0/1 mask的one-hot labels区分）得到新 Logit $s∗cos \theta_j, j\in{1,2,..,y_i,..,n}$。
6、对上述过程得到的**新Logit**按通常方式计算Softmax Loss。
<div align=center>
<img src="./docs/train.jpg"/>
</div>
## 环境配置
### Docker（方法一）
从[光源](https://www.sourcefind.cn/#/service-list)中拉取docker镜像：
```
Note:
It is not recommended to use a single GPU for training, as this may result in longer training times and suboptimal performance. For best results, we suggest using multiple GPUs or a GPU cluster.
### 2. To run on a machine with 8 GPUs:
```shell
torchrun --nproc_per_node=8 train_v2.py configs/ms1mv3_r50
docker pull image.sourcefind.cn:5000/dcu/admin/base/pytorch:1.13.1-centos7.6-dtk23.10-py310
```
### 3. To run on 2 machines with 8 GPUs each:
Node 0:
```shell
torchrun --nproc_per_node=8 --nnodes=2 --node_rank=0 --master_addr="ip1" --master_port=12581 train_v2.py configs/wf42m_pfc02_16gpus_r100
创建容器并挂载目录进行开发：
```
Node 1:
```shell
torchrun --nproc_per_node=8 --nnodes=2 --node_rank=1 --master_addr="ip1" --master_port=12581 train_v2.py configs/wf42m_pfc02_16gpus_r100
docker run -it --name {name} --shm-size=1024G --device=/dev/kfd --device=/dev/dri/ --privileged --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --ulimit memlock=-1:-1 --ipc=host --network host --group-add video -v /opt/hyhal:/opt/hyhal:ro -v {}:{} {docker_image} /bin/bash
# 修改1 {name} 需要改为自定义名称，建议命名{框架_dtk版本_使用者姓名},如果有特殊用途可在命名框架前添加命名
# 修改2 {docker_image} 需要需要创建容器的对应镜像名称，如: pytorch:1.10.0-centos7.6-dtk-23.04-py37-latest【镜像名称:tag名称】
# 修改3 -v 挂载路径到容器指定路径
pip install -r requirements.txt
```
### 4. Run ViT-B on a machine with 24k batchsize:
```shell
torchrun --nproc_per_node=8 train_v2.py configs/wf42m_pfc03_40epoch_8gpu_vit_b
### Dockerfile（方法二）
```
cd docker
docker build --no-cache -t arcface_pytorch:1.0 .
docker run -it --name {name} --shm-size=1024G --device=/dev/kfd --device=/dev/dri/ --privileged --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --ulimit memlock=-1:-1 --ipc=host --network host --group-add video -v /opt/hyhal:/opt/hyhal:ro -v {}:{} {docker_image} /bin/bash
pip install -r requirements.txt
```
### Anaconda（方法三）
线上节点推荐使用conda进行环境配置。
创建python=3.10的conda环境并激活
```
conda create -n arcface python=3.10
conda activate arcface
```
关于本项目DCU显卡所需的特殊深度学习库可从[光合](https://developer.hpccube.com/tool/)开发者社区下载安装。
```
DTK驱动：dtk23.10
python：python3.10
pytorch:1.13.1
torchvision:0.14.1
```
安装其他依赖包
```
pip install -r requirements.txt
```
## 数据集
`MS1MV2\IJBC`
## Download Datasets or Prepare Datasets
- [MS1MV2](https://github.com/deepinsight/insightface/tree/master/recognition/_datasets_#ms1m-arcface-85k-ids58m-images-57) (87k IDs, 5.8M images)
- [MS1MV3](https://github.com/deepinsight/insightface/tree/master/recognition/_datasets_#ms1m-retinaface) (93k IDs, 5.2M images)
- [Glint360K](https://github.com/deepinsight/insightface/tree/master/recognition/partial_fc#4-download) (360k IDs, 17.1M images)
- [WebFace42M](docs/prepare_webface42m.md) (2M IDs, 42.5M images)
- [Your Dataset, Click Here!](docs/prepare_custom_dataset.md)
Note:
If you want to use DALI for data reading, please use the script 'scripts/shuffle_rec.py' to shuffle the InsightFace style rec before using it.
Example:
`python scripts/shuffle_rec.py ms1m-retinaface-t1`
You will get the "shuffled_ms1m-retinaface-t1" folder, where the samples in the "train.rec" file are shuffled.
## Model Zoo
- The models are available for non-commercial research purposes only.
- All models can be found in here.
- [Baidu Yun Pan](https://pan.baidu.com/s/1CL-l4zWqsI1oDuEEYVhj-g): e8pw
- [OneDrive](https://1drv.ms/u/s!AswpsDO2toNKq0lWY69vN58GR6mw?e=p9Ov5d)
### Performance on IJB-C and [**ICCV2021-MFR**](https://github.com/deepinsight/insightface/blob/master/challenges/mfr/README.md)
ICCV2021-MFR testset consists of non-celebrities so we can ensure that it has very few overlap with public available face
recognition training set, such as MS1M and CASIA as they mostly collected from online celebrities.
As the result, we can evaluate the FAIR performance for different algorithms.
For **ICCV2021-MFR-ALL** set, TAR is measured on all-to-all 1:1 protocal, with FAR less than 0.000001(e-6). The
globalised multi-racial testset contains 242,143 identities and 1,624,305 images.
#### 1. Training on Single-Host GPU
| Datasets | Backbone | **MFR-ALL** | IJB-C(1E-4) | IJB-C(1E-5) | log |
|:---------------|:--------------------|:------------|:------------|:------------|:------------------------------------------------------------------------------------------------------------------------------------|
| MS1MV2 | mobilefacenet-0.45G | 62.07 | 93.61 | 90.28 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/ms1mv2_mbf/training.log) |
| MS1MV2 | r50 | 75.13 | 95.97 | 94.07 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/ms1mv2_r50/training.log) |
| MS1MV2 | r100 | 78.12 | 96.37 | 94.27 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/ms1mv2_r100/training.log) |
| MS1MV3 | mobilefacenet-0.45G | 63.78 | 94.23 | 91.33 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/ms1mv3_mbf/training.log) |
| MS1MV3 | r50 | 79.14 | 96.37 | 94.47 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/ms1mv3_r50/training.log) |
| MS1MV3 | r100 | 81.97 | 96.85 | 95.02 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/ms1mv3_r100/training.log) |
| Glint360K | mobilefacenet-0.45G | 70.18 | 95.04 | 92.62 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/glint360k_mbf/training.log) |
| Glint360K | r50 | 86.34 | 97.16 | 95.81 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/glint360k_r50/training.log) |
| Glint360k | r100 | 89.52 | 97.55 | 96.38 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/glint360k_r100/training.log) |
| WF4M | r100 | 89.87 | 97.19 | 95.48 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/wf4m_r100/training.log) |
| WF12M-PFC-0.2 | r100 | 94.75 | 97.60 | 95.90 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/wf12m_pfc02_r100/training.log) |
| WF12M-PFC-0.3 | r100 | 94.71 | 97.64 | 96.01 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/wf12m_pfc03_r100/training.log) |
| WF12M | r100 | 94.69 | 97.59 | 95.97 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/wf12m_r100/training.log) |
| WF42M-PFC-0.2 | r100 | 96.27 | 97.70 | 96.31 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/wf42m_pfc02_r100/training.log) |
| WF42M-PFC-0.2 | ViT-T-1.5G | 92.04 | 97.27 | 95.68 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/wf42m_pfc02_40epoch_8gpu_vit_t/training.log) |
| WF42M-PFC-0.3 | ViT-B-11G | 97.16 | 97.91 | 97.05 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/pfc03_wf42m_vit_b_8gpu/training.log) |
#### 2. Training on Multi-Host GPU
| Datasets | Backbone(bs*gpus) | **MFR-ALL** | IJB-C(1E-4) | IJB-C(1E-5) | Throughout | log |
|:-----------------|:------------------|:------------|:------------|:------------|:-----------|:-------------------------------------------------------------------------------------------------------------------------------------------|
| WF42M-PFC-0.2 | r50(512*8) | 93.83 | 97.53 | 96.16 | ~5900 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/webface42m_r50_bs4k_pfc02/training.log) |
| WF42M-PFC-0.2 | r50(512*16) | 93.96 | 97.46 | 96.12 | ~11000 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/webface42m_r50_lr01_pfc02_bs8k_16gpus/training.log) |
| WF42M-PFC-0.2 | r50(128*32) | 94.04 | 97.48 | 95.94 | ~17000 | click me |
| WF42M-PFC-0.2 | r100(128*16) | 96.28 | 97.80 | 96.57 | ~5200 | click me |
| WF42M-PFC-0.2 | r100(256*16) | 96.69 | 97.85 | 96.63 | ~5200 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/webface42m_r100_bs4k_pfc02/training.log) |
| WF42M-PFC-0.0018 | r100(512*32) | 93.08 | 97.51 | 95.88 | ~10000 | click me |
| WF42M-PFC-0.2 | r100(128*32) | 96.57 | 97.83 | 96.50 | ~9800 | click me |
`r100(128*32)` means backbone is r100, batchsize per gpu is 128, the number of gpus is 32.
#### 3. ViT For Face Recognition
| Datasets | Backbone(bs) | FLOPs | **MFR-ALL** | IJB-C(1E-4) | IJB-C(1E-5) | Throughout | log |
|:--------------|:--------------|:------|:------------|:------------|:------------|:-----------|:-----------------------------------------------------------------------------------------------------------------------------|
| WF42M-PFC-0.3 | r18(128*32) | 2.6 | 79.13 | 95.77 | 93.36 | - | click me |
| WF42M-PFC-0.3 | r50(128*32) | 6.3 | 94.03 | 97.48 | 95.94 | - | click me |
| WF42M-PFC-0.3 | r100(128*32) | 12.1 | 96.69 | 97.82 | 96.45 | - | click me |
| WF42M-PFC-0.3 | r200(128*32) | 23.5 | 97.70 | 97.97 | 96.93 | - | click me |
| WF42M-PFC-0.3 | VIT-T(384*64) | 1.5 | 92.24 | 97.31 | 95.97 | ~35000 | click me |
| WF42M-PFC-0.3 | VIT-S(384*64) | 5.7 | 95.87 | 97.73 | 96.57 | ~25000 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/pfc03_wf42m_vit_s_64gpu/training.log) |
| WF42M-PFC-0.3 | VIT-B(384*64) | 11.4 | 97.42 | 97.90 | 97.04 | ~13800 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/pfc03_wf42m_vit_b_64gpu/training.log) |
| WF42M-PFC-0.3 | VIT-L(384*64) | 25.3 | 97.85 | 98.00 | 97.23 | ~9406 | [click me](https://raw.githubusercontent.com/anxiangsir/insightface_arcface_log/master/pfc03_wf42m_vit_l_64gpu/training.log) |
`WF42M` means WebFace42M, `PFC-0.3` means negivate class centers sample rate is 0.3.
#### 4. Noisy Datasets
| Datasets | Backbone | **MFR-ALL** | IJB-C(1E-4) | IJB-C(1E-5) | log |
|:-------------------------|:---------|:------------|:------------|:------------|:---------|
| WF12M-Flip(40%) | r50 | 43.87 | 88.35 | 80.78 | click me |
| WF12M-Flip(40%)-PFC-0.1* | r50 | 80.20 | 96.11 | 93.79 | click me |
| WF12M-Conflict | r50 | 79.93 | 95.30 | 91.56 | click me |
| WF12M-Conflict-PFC-0.3* | r50 | 91.68 | 97.28 | 95.75 | click me |
`WF12M` means WebFace12M, `+PFC-0.1*` denotes additional abnormal inter-class filtering.
- 训练集[faces_emore.zip](https://pan.baidu.com/s/1S6LJZGdqcZRle1vlcMzHOQ)，
下载后解压到当前目录
数据目录结构如下：
```
── faces_emore
| agedb_30.bin
| calfw.bin
| cfp_ff.bin
| cfp_fp.bin
| cplfw.bin
| lfw.bin
| property
| train.idx
| train.rec
| vgg2_fp.bin
```
- 测试集[IJBC.zip](https://pan.baidu.com/s/1Ok4sqTO8vqAE_kG3zV1rqw?pwd=1234)
解压分卷压缩文件：
```
# 将所有的分卷压缩文件放在一个文件夹中
zip -s 0 IJBC.zip --out IJBC_ALL.zip
unzip IJBC.zip
```
## 训练
Backbone使用ResNet100，在MS1MV3数据集上的预训练权重文件为[model.pt](https://pan.baidu.com/s/1W-TisIZtZmRQz32hq5T6Uw?pwd=1234)
## Speed Benchmark
<div><img src="https://github.com/anxiangsir/insightface_arcface_log/blob/master/pfc_exp.png" width = "90%" /></div>
### 单机单卡
```
python train_v2.py configs/ms1mv2_r100
```
### 单机多卡
```
torchrun --nproc_per_node=4 train_v2.py configs/ms1mv2_r100
```
**Arcface-Torch** is an efficient tool for training large-scale face recognition training sets. When the number of classes in the training sets exceeds one million, the partial FC sampling strategy maintains the same accuracy while providing several times faster training performance and lower GPU memory utilization. The partial FC is a sparse variant of the model parallel architecture for large-scale face recognition, utilizing a sparse softmax that dynamically samples a subset of class centers for each training batch. During each iteration, only a sparse portion of the parameters are updated, leading to a significant reduction in GPU memory requirements and computational demands. With the partial FC approach, it is possible to train sets with up to 29 million identities, the largest to date. Furthermore, the partial FC method supports multi-machine distributed training and mixed precision training.
## 精度
下载权重文件和测试数据集，测试模型精度：
```
python eval_ijbc.py --model-prefix model.pt --image-path IJBC_ALL --network r100
```
模型在MS1MV2数据集的测试指标：
| 模型 | 数据类型 | AUC |
| :------: | :------: | :------: |
| [r34](https://pan.baidu.com/s/1LR0zm8AxwN2tZH55xQdzHw?pwd=1234) | fp16 | 99.4611% |
| [r50](https://pan.baidu.com/s/128GP5J-jWvNbQAAur68bHw?pwd=1234) | fp16 | 99.4854% |
| [r100](https://pan.baidu.com/s/1cslUcKgv5dSrJtBp62J6Fw?pwd=1234) | fp16 | 99.5296% |
| [r100](https://pan.baidu.com/s/1KRBAKFzJU2ZOqhVHe91N6A?pwd=1234) | fp32 | 99.5612% |
## 应用场景
### 算法类别
人脸识别
More details see
[speed_benchmark.md](docs/speed_benchmark.md) in docs.
### 热点应用行业
安防,交通,教育
> 1. Training Speed of Various Parallel Techniques (Samples per Second) on a Tesla V100 32GB x 8 System (Higher is Optimal)
`-` means training failed because of gpu memory limitations.
## 源码仓库及问题反馈
[https://developer.hpccube.com/codes/modelzoo/arcface_pytorch](https://developer.hpccube.com/codes/modelzoo/arcface_pytorch)
## 参考资料
[https://github.com/deepinsight/insightface/tree/master/recognition/arcface_torch](https://github.com/deepinsight/insightface/tree/master/recognition/arcface_torch)
| Number of Identities in Dataset | Data Parallel | Model Parallel | Partial FC 0.1 |
|:--------------------------------|:--------------|:---------------|:---------------|
| 125000 | 4681 | 4824 | 5004 |
| 1400000 | **1672** | 3043 | 4738 |
| 5500000 | **-** | **1389** | 3975 |
| 8000000 | **-** | **-** | 3565 |
| 16000000 | **-** | **-** | 2679 |
| 29000000 | **-** | **-** | **1855** |
> 2. GPU Memory Utilization of Various Parallel Techniques (MB per GPU) on a Tesla V100 32GB x 8 System (Lower is Optimal)
| Number of Identities in Dataset | Data Parallel | Model Parallel | Partial FC 0.1 |
|:--------------------------------|:--------------|:---------------|:---------------|
| 125000 | 7358 | 5306 | 4868 |
| 1400000 | 32252 | 11178 | 6056 |
| 5500000 | **-** | 32188 | 9854 |
| 8000000 | **-** | **-** | 12310 |
| 16000000 | **-** | **-** | 19950 |
| 29000000 | **-** | **-** | 32324 |
## Citations
```
@inproceedings{deng2019arcface,
title={Arcface: Additive angular margin loss for deep face recognition},
author={Deng, Jiankang and Guo, Jia and Xue, Niannan and Zafeiriou, Stefanos},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
pages={4690--4699},
year={2019}
}
@inproceedings{An_2022_CVPR,
author={An, Xiang and Deng, Jiankang and Guo, Jia and Feng, Ziyong and Zhu, XuHan and Yang, Jing and Liu, Tongliang},
title={Killing Two Birds With One Stone: Efficient and Robust Training of Face Recognition CNNs by Partial FC},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month={June},
year={2022},
pages={4042-4051}
}
@inproceedings{zhu2021webface260m,
title={Webface260m: A benchmark unveiling the power of million-scale deep face recognition},
author={Zhu, Zheng and Huang, Guan and Deng, Jiankang and Ye, Yun and Huang, Junjie and Chen, Xinze and Zhu, Jiagang and Yang, Tian and Lu, Jiwen and Du, Dalong and Zhou, Jie},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={10492--10502},
year={2021}
}
```
README2.md deleted 100644 → 0
View file @ 9b87ce51
# ArcFace
## 论文
- https://arxiv.org/pdf/1801.07698.pdf
## 模型结构
这篇文章提出一种新的用于人脸识别的损失函数：additive angular margin loss，直接在角度空间（angular space）中最大化分类界限，基于该损失函数训练得到人脸识别算法ArcFace。
<div align=center>
<img src="./docs/arcface.png"/>
</div>
## 算法原理
通过训练深度卷积神经网络嵌入 (DCNN Embedding) 来进行人脸识别。
ArcFace训练流程：
设类别数（人脸ID数量）为 $n$,DCNN的最后一个FC 层的权重为$W\subset {\mathbb{R}}^{d \times n}$，输入$W$的特征$x_i$的维度为$d$。
1、分别归一化输入特征$x_i \subset {\mathbb{R}}^{b}$和FC层权重$W_j \in {\mathbb{R}}^{1 \times b}$（张量除以欧几里得范数标量），令所得归一化特征$\frac{x_i}{\|x_i\|}$与第$j \in {1,2,...,y_i,...,n}$个类别的FC层权重$\frac{{W_j}^T}{\|W_j\|} \in {\mathbb{R}}^{1\times d}$点乘得到FC层的第$j$个输出$cos \theta_j \in {\mathbb{R}}^{1\times1}$（数量积公式：${W_j}^{T}\cdot x_i=\|W_j\|\|x_i\|cos\theta_j$）,表示**将特征$x_i$预测为第$j$类的预测值**。
2、设特征$x_i$的真实类别为第$y_j$个类别，单独取出Target权重$\frac{{W_{y_j}}^T}{\|W_{y_i}\|}$计算$\theta_{y_i}=arccos(cos\theta_{y_i})=arccos(\frac{{W_{y_j}}^T}{\|W_{y_i}\|}\cdot\frac{x_i}{\|x_i\|})$可得归一化特征$\frac{x_i}{\|x_i\|}$与归一化**target权重**$\frac{{W_{y_j}}^T}{\|W_{y_i}\|}$之间的夹角—— **Target角度$\theta_{y_i}$**
3、通过把一个自定义的**加性角度边距 (additive angular margin)** $m$加到$\theta_{y_i}$，得到$\theta_{y_i}+m$，用于**调整Target角度**。
4、计算经调整的Target角度的余弦，得到仅关于特征$x_i$的真实类别$y_i$的**新Target Logit $cos(\theta_{y_i}+m)$**。
5、通过自定义的特征范数$s$重缩放所有Logit（除Target Logit变为$cos(\theta_{y_i}+m)$）外其余原Logit仍为$cos\theta_j$，矩阵运算时需用相当于 0/1 mask的one-hot labels区分）得到新 Logit $s∗cos \theta_j, j\in{1,2,..,y_i,..,n}$。
6、对上述过程得到的**新Logit**按通常方式计算Softmax Loss。
<div align=center>
<img src="./docs/train.jpg"/>
</div>
## 环境配置
### Docker（方法一）
从[光源](https://www.sourcefind.cn/#/service-list)中拉取docker镜像：
```
docker pull image.sourcefind.cn:5000/dcu/admin/base/pytorch:1.13.1-centos7.6-dtk23.10-py310
```
创建容器并挂载目录进行开发：
```
docker run -it --name {name} --shm-size=1024G --device=/dev/kfd --device=/dev/dri/ --privileged --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --ulimit memlock=-1:-1 --ipc=host --network host --group-add video -v /opt/hyhal:/opt/hyhal:ro -v {}:{} {docker_image} /bin/bash
# 修改1 {name} 需要改为自定义名称，建议命名{框架_dtk版本_使用者姓名},如果有特殊用途可在命名框架前添加命名
# 修改2 {docker_image} 需要需要创建容器的对应镜像名称，如: pytorch:1.10.0-centos7.6-dtk-23.04-py37-latest【镜像名称:tag名称】
# 修改3 -v 挂载路径到容器指定路径
pip install -r requirements.txt
```
### Dockerfile（方法二）
```
cd docker
docker build --no-cache -t arcface_pytorch:1.0 .
docker run -it --name {name} --shm-size=1024G --device=/dev/kfd --device=/dev/dri/ --privileged --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --ulimit memlock=-1:-1 --ipc=host --network host --group-add video -v /opt/hyhal:/opt/hyhal:ro -v {}:{} {docker_image} /bin/bash
pip install -r requirements.txt
```
### Anaconda（方法三）
线上节点推荐使用conda进行环境配置。
创建python=3.10的conda环境并激活
```
conda create -n arcface python=3.10
conda activate arcface
```
关于本项目DCU显卡所需的特殊深度学习库可从[光合](https://developer.hpccube.com/tool/)开发者社区下载安装。
```
DTK驱动：dtk23.10
python：python3.10
pytorch:1.13.1
torchvision:0.14.1
```
安装其他依赖包
```
pip install -r requirements.txt
```
## 数据集
`MS1MV2`
- 训练集[face_train.zip](https://pan.baidu.com/s/1S6LJZGdqcZRle1vlcMzHOQ)，
下载后解压到当前目录
数据目录结构如下：
```
── faces_emore
| agedb_30.bin
| calfw.bin
| cfp_ff.bin
| cfp_fp.bin
| cplfw.bin
| lfw.bin
| property
| train.idx
| train.rec
| vgg2_fp.bin
```
- 测试集[face_val.zip]()
- 项目中已提供用于试验训练的迷你数据集[datasets](https://pan.baidu.com/s/1oKRgOW7jCLxzPZoQofl1mQ?pwd=0okl)，下载后解压即可。
## 训练
Backbone使用ResNet100，在MS1MV3数据机上的预训练权重文件[model.pt](https://pan.baidu.com/s/1W-TisIZtZmRQz32hq5T6Uw?pwd=1234)
### 单机单卡
```
python train_v2.py configs/ms1mv2_r100
```
### 单机多卡
```
torchrun --nproc_per_node=4 train_v2.py configs/ms1mv2_r100
```
## 精度
使用权重文件[model.pt](https://pan.baidu.com/s/1W-TisIZtZmRQz32hq5T6Uw?pwd=1234)，测试模型精度：
```
python eval_ijbc.py --model-prefix model.pt --image-path IJB_release/IJBC/ --network r100
```
### 精度
测试数据：[test](http://images.cocodataset.org/zips/test2017.zip)
测试指标：
| 模型 | 数据类型 | map0.5:0.95 | map0.5 |
| :------: | :------: | :------: | :------: |
| yolo9-c-converted | 全精度 | 0.530 | 0.703 |
| yolo9-e-converted | 全精度 | 0.556 | 0.728 |
| yolo9-c | 全精度 | 0.530 | 0.703 |
| yolo9-e | 全精度 | 0.556 | 0.728 |
| gelan-c | 全精度 | 0.526 | 0.695 |
| gelan-e | 全精度 | 0.550 | 0.719 |
## 应用场景
### 算法类别
人脸识别
### 热点应用行业
安防,交通,教育
## 源码仓库及问题反馈
[https://developer.hpccube.com/codes/modelzoo/yolov9_pytorch](https://developer.hpccube.com/codes/modelzoo/yolov9_pytorch)
## 参考资料
[https://github.com/deepinsight/insightface/tree/master/recognition/arcface_torch](https://github.com/deepinsight/insightface/tree/master/recognition/arcface_torch)
configs/ms1mv2_r100.py
View file @ 8760acf2
......@@ -19,7 +19,7 @@ config.lr = 0.1
config.verbose = 2000
config.dali = False
config.rec = "/train_tmp/faces_emore"
config.rec = "faces_emore"
config.num_classes = 85742
config.num_image = 5822653
config.num_epoch = 20
......
image.png 0 → 100644
View file @ 8760acf2
image.png

65.8 KB

model.properties 0 → 100644
View file @ 8760acf2
# 模型唯一标识
modelCode=676
# 模型名称
modelName=arcface_pytorch
# 模型描述
modelDescription=ArcFace设计了新的用于人脸识别的损失函数,
# 应用场景
appScenario=推理,训练,人脸识别,安防,交通,教育
# 框架类型
frameType=pytorch
requirement.txt
View file @ 8760acf2
......@@ -2,5 +2,8 @@ tensorboard
easydict
mxnet
onnx
sklearn
opencv-python
\ No newline at end of file
scipy==1.7.1
scikit-learn
opencv-python
menpo
prettytable
\ No newline at end of file
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment