Files
insightface/challenges/iccv21-mfr/README.md
2021-10-26 12:23:32 +08:00

186 lines
10 KiB
Markdown

# InsightFace Track of ICCV21-MFR
## NEWS
**``2021-10-26``** Please send the onnx models to us(insightface.challenge[at]gmail.com) if you want to test the MFR accuracy before our system rebooting(may be in Nov.).
**``2021-10-11``** [Final Leaderboard](https://insightface.ai/mfr21)
**``2021-10-04``** Please fix the public leaderboard scores before ``2021-10-05 20:00(UTC+8 Time)``
**``2021-07-16``** Implicit batch inference is prohibited. For example, inserting some data-related OPs to onnx graph to enable automatic flip-test is not allowed(or similar ideas). We will check it after submission closed, to ensure fairness.
**``2021-06-17``** Participants are now ordered in terms of highest scores across two datasets: **TAR@Mask** and **TAR@MR-All**, by the formula of ``0.25 * TAR@Mask + 0.75 * TAR@MR-All``.
## Introduction
The Masked Face Recognition Challenge & Workshop(MFR) will be held in conjunction with the International Conference on Computer Vision (ICCV) 2021.
[Workshop-Homepage](https://ibug.doc.ic.ac.uk/resources/masked-face-recognition-challenge-workshop-iccv-21/).
There're InsightFace track here and [Webface260M](https://www.face-benchmark.org/challenge.html) track(with larger training set) in this workshop.
[**Challenge Leaderboard**](https://insightface.ai/mfr21)
Submission server link: [http://iccv21-mfr.com/](http://iccv21-mfr.com/)
An alternative submission server for Non-Chinese users: [http://124.156.136.55/](http://124.156.136.55/)
### Discussion group:
WeChat:
<img src="https://github.com/nttstar/insightface-resources/blob/master/images/mfr_wechat_group.png" alt="mfr_group" width="360">
QQ Group: 711302608, *answer: github*
Online issue discussion: [https://github.com/deepinsight/insightface/issues/1564](https://github.com/deepinsight/insightface/issues/1564)
## Testsets for insightface track
In this challenge, we will evaluate the accuracy of following testsets:
* Accuracy between masked and non-masked faces.
* Accuracy among children(2~16 years old).
* Accuracy of globalised multi-racial benchmarks.
We ensure that there's no overlap between these testsets and public available training datasets, as they are not collected from online celebrities.
Our test datasets mainly comes from [IFRT](../ifrt).
### ``Mask test-set:``
Mask testset contains 6,964 identities, 6,964 masked images and 13,928 non-masked images. There are totally 13,928 positive pairs and 96,983,824 negative pairs.
<details>
<summary>Click to check the sample images(here we manually blur it to protect privacy) </summary>
<img src="https://github.com/nttstar/insightface-resources/blob/master/images/ifrt_mask_sample.jpg" alt="ifrtsample" width="360">
</details>
### ``Children test-set:``
Children testset contains 14,344 identities and 157,280 images. There are totally 1,773,428 positive pairs and 24,735,067,692 negative pairs.
<details>
<summary>Click to check the sample images(here we manually blur it to protect privacy) </summary>
<img src="https://github.com/nttstar/insightface-resources/blob/master/images/ifrt_children_sample.jpg" alt="ifrtsample" width="360">
</details>
### ``Multi-racial test-set (MR in short):``
The globalised multi-racial testset contains 242,143 identities and 1,624,305 images.
| Race-Set | Identities | Images | Positive Pairs | Negative Pairs |
| ------- | ---------- | ----------- | ----------- | ----------- |
| African | 43,874 | 298,010 | 870,091 | 88,808,791,999 |
| Caucasian | 103,293 | 697,245 | 2,024,609 | 486,147,868,171 |
| Indian | 35,086 | 237,080 | 688,259 | 56,206,001,061 |
| Asian | 59,890 | 391,970 | 1,106,078 | 153,638,982,852 |
| **ALL** | **242,143** | **1,624,305** | **4,689,037** | **2,638,360,419,683** |
<details>
<summary>Click to check the sample images(here we manually blur it to protect privacy) </summary>
<img src="https://github.com/nttstar/insightface-resources/blob/master/images/ifrtsample_blur.jpg" alt="ifrtsample" width="640">
</details>
## Evaluation Metric
For ``Mask`` set, TAR is measured on mask-to-nonmask 1:1 protocal, with FAR less than 0.0001(e-4).
For ``Children`` set, TAR is measured on all-to-all 1:1 protocal, with FAR less than 0.0001(e-4).
For other sets, TAR is measured on all-to-all 1:1 protocal, with FAR less than 0.000001(e-6).
Participants are ordered in terms of highest scores across two datasets: **TAR@Mask** and **TAR@MR-All**, by the formula of ``0.25 * TAR@Mask + 0.75 * TAR@MR-All``.
## Baselines
| Backbone | Dataset | Method | Mask | Children | African | Caucasian | South Asian | East Asian | MR-All | size(mb) | infer(ms) | link |
|------------|------------|------------|--------|----------|---------|-----------|-------------|------------|--------|----------|-----------|-----------|
| R100 | Casia | ArcFace | 26.623 | 30.359 | 39.666 | 53.933 | 47.807 | 21.572 | 42.735 | 248.904 | 7.073 | [download](https://1drv.ms/u/s!AswpsDO2toNKrUJpk8zC61HVN7Kg?e=zE9JDd) |
| R100 | MS1MV2 | ArcFace | 65.767 | 60.496 | 79.117 | 87.176 | 85.501 | 55.807 | 80.725 | 248.904 | 7.028 | [download](https://1drv.ms/u/s!AswpsDO2toNKrUTlYEHJCHg3UYM-?e=ihxMpS) |
| R18 | MS1MV3 | ArcFace | 47.853 | 41.047 | 62.613 | 75.125 | 70.213 | 43.859 | 68.326 | 91.658 | 1.856 | [download](https://1drv.ms/u/s!AswpsDO2toNKrTxlT6w1Jo02yzSh?e=KDhFAA) |
| R34 | MS1MV3 | ArcFace | 58.723 | 55.834 | 71.644 | 83.291 | 80.084 | 53.712 | 77.365 | 130.245 | 3.054 | [download](https://1drv.ms/u/s!AswpsDO2toNKrT2O5pgyVtwnjeMq?e=16S8LI) |
| R50 | MS1MV3 | ArcFace | 63.850 | 60.457 | 75.488 | 86.115 | 84.305 | 57.352 | 80.533 | 166.305 | 4.262 | [download](https://1drv.ms/u/s!AswpsDO2toNKrUUWd5i3a5OlFpM_?e=ExBDBN) |
| R100 | MS1MV3 | ArcFace | 69.091 | 66.864 | 81.083 | 89.040 | 88.082 | 62.193 | 84.312 | 248.590 | 7.031 | [download](https://1drv.ms/u/s!AswpsDO2toNKrUPwyqWvNXUlNd3P?e=pTLw9A) |
| R18 | Glint360K | ArcFace | 53.317 | 48.113 | 68.230 | 80.575 | 75.852 | 47.831 | 72.074 | 91.658 | 2.013 | [download](https://1drv.ms/u/s!AswpsDO2toNKrT5ey4lCqFzlpzDd?e=VWP28J) |
| R34 | Glint360K | ArcFace | 65.106 | 65.454 | 79.907 | 88.620 | 86.815 | 60.604 | 83.015 | 130.245 | 3.044 | [download](https://1drv.ms/u/s!AswpsDO2toNKrUBcgGkiuUS11Hsd?e=ISGDnP) |
| R50 | Glint360K | ArcFace | 70.233 | 69.952 | 85.272 | 91.617 | 90.541 | 66.813 | 87.077 | 166.305 | 4.340 | [download](https://1drv.ms/u/s!AswpsDO2toNKrT8jbvHxjqCY0d08?e=igfdrd) |
| R100 | Glint360K | ArcFace | 75.567 | 75.202 | 89.488 | 94.285 | 93.434 | 72.528 | 90.659 | 248.590 | 7.038 | [download](https://1drv.ms/u/s!AswpsDO2toNKrUFgLEIj-mnkb51b?e=vWqy2q) |
| - | *Private* | <div style="width: 50pt">insightface-000 of frvt | 97.760 | 93.358 | 98.850 | 99.372 | 99.058 | 87.694 | 97.481 | - | - | - |
(MS1M-V2 means MS1M-ArcFace, MS1M-V3 means MS1M-RetinaFace).
Inference time was evaluated on Tesla V100 GPU, using onnxruntime-gpu==1.6.
## Rules
1. We have two sub-tracks, determined by the size of training dataset and inference time limitation.
* Sub-Track A: Use MS1M-V3 as training set, download: [ref-link](https://github.com/deepinsight/insightface/blob/master/recognition/_datasets_/README.md#ms1m-retinaface), feature length must <= 512, and the inference time must <= 10ms on Tesla V100 GPU.
* Sub-Track B: Use Glint360K as training set, download: [ref-link](https://github.com/deepinsight/insightface/blob/master/recognition/_datasets_/README.md#deepglint-181k-ids675m-images-8), feature length must <= 1024, and the inference time must <= 20ms on Tesla V100 GPU.
2. Training set and testing set are both aligned to 112x112, re-alignment is prohibited.
3. Mask data-augmentation is allowed, such as [this](../../recognition/_tools_). The applied mask augmentation tool should be reproducible.
4. External dataset and pretrained models are both prohibited.
5. Participants submit onnx model, then get scores by our online evaluation. Test images are invisible.
6. Matching score is measured by cosine similarity.
7. Model size must <= 1GB.
8. The input shape of submission model should equal to 3x112x112 (RGB order).
9. Online evaluation server uses onnxruntime-gpu==1.6, cuda==10.2, cudnn==8.0.5.
10. Any float-16 model weights is prohibited, as it will lead to incorrect model size estimiation.
11. Please use ``onnx_helper.py`` to check whether the model is valid.
12. Participants are now ordered in terms of highest scores across two datasets: **TAR@Mask** and **TAR@MR-All**, by the formula of ``0.25 * TAR@Mask + 0.75 * TAR@MR-All``.
13. Top-ranked participants should provide their solutions and codes to ensure their validity after submission closed.
## Tutorial
1. ArcFace-PyTorch (with Partial-FC), [code](../../recognition/arcface_torch), [tutorial-cn](tutorial_pytorch_cn.md)
2. OneFlow, [code](../../recognition/oneflow_face)
3. MXNet, [code](../../recognition/arcface_mxnet)
## Submission Guide
1. Participants must package the onnx model for submission using ``zip xxx.zip model.onnx``.
2. Each participant can submit three times a day at most.
3. Please sign-up with the real organization name. You can hide the organization name in our system if you like.
4. You can decide which submission to be displayed on the leaderboard by clicking 'Set Public' button.
5. Please click 'sign-in' on submission server if find you're not logged in.
Server link: [http://iccv21-mfr.com/](http://iccv21-mfr.com/)
## Timelines
* 1 June - Release of the training data, baseline solutions and testing leader-board
* 1 October - Stop leader-board submission (11:59 PM Pacific Time)
* 7 October - Winners notification
## Sponsors
(in alphabetical order)
**[DeepGlint](http://www.deepglint.com/)**
**[Kiwi Tech](http://www.kiwiar.com)**
**[OneFlow](https://www.oneflow.org)**
**[Zoloz](https://www.zoloz.com)**
## Bonus Share
| | Sub-Track A | Sub-Track B |
| --------- | ------- | ------- |
| 1st place | 30% | 30% |
| 2nd place | 15% | 15% |
| 3rd place | 5% | 5% |