Algorithm | Date Submitted | Set 1 | Set 2 | Set 3 | Data Set Size |
---|---|---|---|---|---|
BingMMLab V1(iBUG cleaned data) | 4/10/2018 | 98.998% | 98.998% | 98.998% | Large |
Orion Star Technology (clean) | 3/21/2018 | 98.355% | Large | ||
iBUG_DeepInsight | 2/8/2018 | 98.063% | 98.058% | 98.053% | Large |
EM-DATA | 4/4/2018 | 96.653% | 96.653% | 96.653% | Large |
SuningUS_AILab | 3/21/2018 | 96.2618947% | 96.2618947% | 96.2618947% | Large |
StartDT-AI | 4/16/2018 | 93.8226% | 93.8226% | 93.8226% | Large |
Intellivision | 2/11/2018 | 93.125% | 93.123% | 93.136% | Large |
ULSee - Face Team | 3/27/2018 | 92.172% | Large | ||
Vocord - deepVo V3 | 04/27/2017 | 91.763% | 91.711% | 91.704% | Large |
MTDP_ITC | 12/21/2017 | 87.098% | 83.877% | 87.184% | Large |
TUPUTECH | 12/22/2017 | 86.558% | 86.557% | 86.579% | Large |
Video++ | 1/5/2018 | 85.74% | 85.737% | 85.735% | Large |
THU CV-AI Lab | 12/12/2017 | 84.521% | 84.513% | 84.514% | Large |
TencentAILab_FaceCNN_v1 | 9/21/2017 | 84.261% | 84.255% | 84.257% | Large |
BingMMLab-v1 (non-cleaned data) | 4/10/2018 | 83.758% | 83.758% | 83.758% | Large |
Orion Star Technology (no clean) | 3/21/2018 | 83.569% | Large | ||
YouTu Lab (Tencent Best-Image) | 04/08/2017 | 83.29% | 83.267% | 83.295% | Large |
OceanAI Tech | 12/15/2017 | 83.164% | 83.177% | 83.144% | Large |
FaceTag V1 | 12/18/2017 | 82.411% | 82.376% | 82.364% | Large |
Argus_v1 | 2/6/2018 | 81.437% | 81.424% | 81.438% | Large |
Yang Sun | 06/05/2017 | 81.326% | 81.286% | 81.284% | Large |
DeepSense V2 | 1/22/2017 | 81.298% | 81.298% | 81.298% | Large |
iBug (Reported by Author) | 04/28/2017 | 80.277% | Small | ||
Vocord-deepVo1.2 | 12/1/2016 | 80.258% | 80.195% | 80.241% | Large |
Faceter Lab | 12/18/2017 | 79.426% | 79.367% | 79.356% | Large |
Progressor | 09/13/2017 | 79.41% | 79.41% | 79.41% | Large |
CVTE V2 | 1/27/2018 | 78.324% | 78.293% | 78.298% | Small |
Fudan University - FUDAN-CS_SDS | 1/29/2017 | 77.982% | 78.006% | 77.99% | Small |
GRCCV | 12/1/2016 | 77.677% | 77.021% | 77.147% | Small |
XT-tech V2 | 08/30/2017 | 77.239% | 77.239% | 77.239% | Large |
Beijing Faceall Co. - FaceAll V2 | 04/28/2017 | 76.661% | 76.643% | 76.607% | Small |
SphereFace - Small | 12/1/2016 | 75.766% | 75.765% | 75.77% | Small |
Vocord - DeepVo1 | 08/03/2016 | 75.127% | 75.093% | 75.125% | Large |
DeepSense - Large | 07/31/2016 | 74.799% | 74.78% | 74.813% | Large |
SIATMMLAB TencentVision | 12/1/2016 | 74.207% | 74.213% | 74.195% | Large |
Shanghai Tech | 08/13/2016 | 74.049% | 74.032% | 74.02% | Large |
CVTE | 08/30/2017 | 73.521% | 73.501% | 73.504% | Large |
NTechLAB - facenx_large | 10/20/2015 | 73.3% | 73.309% | 73.287% | Large |
ForceInfo | 04/07/2017 | 72.11% | 72.084% | 72.121% | Large |
3DiVi Company - tdvm V2 | 04/15/2017 | 71.742% | 71.727% | 71.703% | Large |
DeepSense - Small | 07/31/2016 | 70.983% | 70.948% | 70.962% | Small |
Google - FaceNet v8 | 10/23/2015 | 70.496% | 70.492% | 70.551% | Large |
SIAT_MMLAB | 3/29/2016 | 65.233% | 65.223% | 65.229% | Small |
Beijing Faceall Co. - FaceAll_Norm_1600 | 10/19/2015 | 64.803% | 64.798% | 64.826% | Large |
Beijing Faceall Co. - FaceAll_1600 | 10/19/2015 | 63.977% | 63.962% | 63.993% | Large |
Barebones_FR - cnn | 10/21/2015 | 59.363% | 59.379% | 59.389% | Small |
NTechLAB - facenx_small | 10/20/2015 | 58.218% | 58.208% | 58.21% | Small |
3DiVi Company - tdvm6 | 10/27/2015 | 33.705% | 33.69% | 33.667% | Small |
Joint Bayes | 10/20/2015 | 3.021% | 3.223% | 3.245% | Small |
LBP | 10/20/2015 | 2.326% | 2.32% | 2.318% | Small |
Algorithm | Details |
---|---|
Orion Star Technology (clean) | We have trained three deep networks (ResNet-101, ResNet-152, ResNet-200) with joint softmax and triplet loss on MS-Celeb-1M (95K identities, 5.1M images), and the triplet part is trained by batch online hard negative mining with subspace learning. The features of all networks are concatenated to produce the final feature, whose dimension is set to be 256x3. For data processing, we use original large images and follow our own system by detection and alignment. Particularly, in evaluation, we have cleaned the FaceScrub and MegaFace with the code released by iBUG_DeepInsight. |
Orion Star Technology (no clean) | Compared to our another submission named “Orion Star Technology (clean)”, the major difference is that no any data cleaning is adopted in evaluation. |
SuningUS_AILab | This is a model ensembled by three different models using ResNet CNN and improved ResNet network, learned by a combined loss. A filtered MS-Celeb-1M and CASIA-Webface is used as the dataset. |
ULSee - Face Team | Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks https://arxiv.org/abs/1604.02878 Large-Margin Softmax Loss for Convolutional Neural Networks https://arxiv.org/abs/1612.02295 A Discriminative Deep Feature Learning Approach for Face Recognition https://ydwen.github.io/papers/WenECCV16.pdf NormFace: L2 Hypersphere Embedding for Face Verification https://arxiv.org/abs/1704.06369 SphereFace: Deep Hypersphere Embedding for Face Recognition https://arxiv.org/abs/1704.08063 ArcFace: Additive Angular Margin Loss for Deep Face Recognition https://arxiv.org/abs/1801.07698 |
SIATMMLAB TencentVision | Adopt the ensemble of very deep CNNs, learned by joint supervision (softmax loss, improved center loss, etc). Training data is a combination of public datasets (CAISA, VGG, CACD2000, etc) and private datasets. The total number of images is more than 2 million. |
DeepSense - Small | Adopt a network of very deep ResNet CNNs, learned by combined supervision(identification loss(softmax loss), verification loss, triplet loss). |
GRCCV | The algorithm consists of three parts: FCN - based fast face detection algorithm, pre-training ResNet CNN on classification task, weight tuning. Training set contains 273000 photos. Hardware: 8 x Tesla k80. |
SphereFace - Small | SphereFace uses a novel approach to learn face features that are discriminative on a hypersphere manifold. The training data set we use in SphereFace is the publicly available CASIA-WebFace dataset which contains 490k images of nearly 10,500 individuals. |
EM-DATA | arcface, https://github.com/deepinsight/insightface |
StartDT-AI | we only use a single model trained on a ResNet-28 network joined with cosine loss and triplet loss on MS-Celeb-1M(74K identities, 4M images), refer to DeepVisage: Making face recognition simple yet with powerful generalization skills https://arxiv.org/abs/1703.08388 One-shot Face Recognition by Promoting Underrepresented Classes https://arxiv.org/abs/1707.05574v2 ArcFace: Additive Angular Margin Loss for Deep Face Recognition https://arxiv.org/abs/1801.07698 SphereFace: Deep Hypersphere Embedding for Face Recognition https://arxiv.org/abs/1704.08063 |
BingMMLab-v1 (non-cleaned data) | Compares to our submissions named “BingMMLab V1(cleaned data)”, the only difference is that no data cleaning is adopted in this evaluation. |
BingMMLab V1(iBUG cleaned data) | We used knowledge graph to collect identities and then crawled Bing search engine to get high quality images, we filtered noises in 14M training data by clustering with weak face models, and then trained a DensetNet-69 (k=48) network with A-Softmax loss variants. In evaluation, we used the cleaned test set released by iBUG_DeepInsight. DenseNet https://arxiv.org/pdf/1608.06993v2.pdf A-Softmax and its variant: https://arxiv.org/pdf/1704.08063.pdf https://github.com/wy1iu/LargeMargin_Softmax_Loss/issues/13 |
- -   uses large training set
- -   uses large training set
- -   uses large training set
Algorithm | Date Submitted | Set 1 | Set 2 | Set 3 | Data Set Size |
---|---|---|---|---|---|
BingMMLab V1(iBUG cleaned data) | 4/10/2018 | 98.998% | 98.998% | 98.998% | Large |
Orion Star Technology (clean) | 3/21/2018 | 98.355% | Large | ||
iBUG_DeepInsight | 2/8/2018 | 98.063% | 98.058% | 98.053% | Large |
EM-DATA | 4/4/2018 | 96.653% | 96.653% | 96.653% | Large |
SuningUS_AILab | 3/21/2018 | 96.2618947% | 96.2618947% | 96.2618947% | Large |
StartDT-AI | 4/16/2018 | 93.8226% | 93.8226% | 93.8226% | Large |
Intellivision | 2/11/2018 | 93.125% | 93.123% | 93.136% | Large |
ULSee - Face Team | 3/27/2018 | 92.172% | Large | ||
Vocord - deepVo V3 | 04/27/2017 | 91.763% | 91.711% | 91.704% | Large |
MTDP_ITC | 12/21/2017 | 87.098% | 83.877% | 87.184% | Large |
TUPUTECH | 12/22/2017 | 86.558% | 86.557% | 86.579% | Large |
Video++ | 1/5/2018 | 85.74% | 85.737% | 85.735% | Large |
THU CV-AI Lab | 12/12/2017 | 84.521% | 84.513% | 84.514% | Large |
TencentAILab_FaceCNN_v1 | 9/21/2017 | 84.261% | 84.255% | 84.257% | Large |
BingMMLab-v1 (non-cleaned data) | 4/10/2018 | 83.758% | 83.758% | 83.758% | Large |
Orion Star Technology (no clean) | 3/21/2018 | 83.569% | Large | ||
YouTu Lab (Tencent Best-Image) | 04/08/2017 | 83.29% | 83.267% | 83.295% | Large |
OceanAI Tech | 12/15/2017 | 83.164% | 83.177% | 83.144% | Large |
FaceTag V1 | 12/18/2017 | 82.411% | 82.376% | 82.364% | Large |
Argus_v1 | 2/6/2018 | 81.437% | 81.424% | 81.438% | Large |
Yang Sun | 06/05/2017 | 81.326% | 81.286% | 81.284% | Large |
DeepSense V2 | 1/22/2017 | 81.298% | 81.298% | 81.298% | Large |
iBug (Reported by Author) | 04/28/2017 | 80.277% | Small | ||
Vocord-deepVo1.2 | 12/1/2016 | 80.258% | 80.195% | 80.241% | Large |
Faceter Lab | 12/18/2017 | 79.426% | 79.367% | 79.356% | Large |
Progressor | 09/13/2017 | 79.41% | 79.41% | 79.41% | Large |
CVTE V2 | 1/27/2018 | 78.324% | 78.293% | 78.298% | Small |
Fudan University - FUDAN-CS_SDS | 1/29/2017 | 77.982% | 78.006% | 77.99% | Small |
GRCCV | 12/1/2016 | 77.677% | 77.021% | 77.147% | Small |
XT-tech V2 | 08/30/2017 | 77.239% | 77.239% | 77.239% | Large |
Beijing Faceall Co. - FaceAll V2 | 04/28/2017 | 76.661% | 76.643% | 76.607% | Small |
SphereFace - Small | 12/1/2016 | 75.766% | 75.765% | 75.77% | Small |
Vocord - DeepVo1 | 08/03/2016 | 75.127% | 75.093% | 75.125% | Large |
DeepSense - Large | 07/31/2016 | 74.799% | 74.78% | 74.813% | Large |
SIATMMLAB TencentVision | 12/1/2016 | 74.207% | 74.213% | 74.195% | Large |
Shanghai Tech | 08/13/2016 | 74.049% | 74.032% | 74.02% | Large |
CVTE | 08/30/2017 | 73.521% | 73.501% | 73.504% | Large |
NTechLAB - facenx_large | 10/20/2015 | 73.3% | 73.309% | 73.287% | Large |
ForceInfo | 04/07/2017 | 72.11% | 72.084% | 72.121% | Large |
3DiVi Company - tdvm V2 | 04/15/2017 | 71.742% | 71.727% | 71.703% | Large |
DeepSense - Small | 07/31/2016 | 70.983% | 70.948% | 70.962% | Small |
Google - FaceNet v8 | 10/23/2015 | 70.496% | 70.492% | 70.551% | Large |
SIAT_MMLAB | 3/29/2016 | 65.233% | 65.223% | 65.229% | Small |
Beijing Faceall Co. - FaceAll_Norm_1600 | 10/19/2015 | 64.803% | 64.798% | 64.826% | Large |
Beijing Faceall Co. - FaceAll_1600 | 10/19/2015 | 63.977% | 63.962% | 63.993% | Large |
Barebones_FR - cnn | 10/21/2015 | 59.363% | 59.379% | 59.389% | Small |
NTechLAB - facenx_small | 10/20/2015 | 58.218% | 58.208% | 58.21% | Small |
3DiVi Company - tdvm6 | 10/27/2015 | 33.705% | 33.69% | 33.667% | Small |
Joint Bayes | 10/20/2015 | 3.021% | 3.223% | 3.245% | Small |
LBP | 10/20/2015 | 2.326% | 2.32% | 2.318% | Small |
Algorithm | Details |
---|---|
Orion Star Technology (clean) | We have trained three deep networks (ResNet-101, ResNet-152, ResNet-200) with joint softmax and triplet loss on MS-Celeb-1M (95K identities, 5.1M images), and the triplet part is trained by batch online hard negative mining with subspace learning. The features of all networks are concatenated to produce the final feature, whose dimension is set to be 256x3. For data processing, we use original large images and follow our own system by detection and alignment. Particularly, in evaluation, we have cleaned the FaceScrub and MegaFace with the code released by iBUG_DeepInsight. |
Orion Star Technology (no clean) | Compared to our another submission named “Orion Star Technology (clean)”, the major difference is that no any data cleaning is adopted in evaluation. |
SuningUS_AILab | This is a model ensembled by three different models using ResNet CNN and improved ResNet network, learned by a combined loss. A filtered MS-Celeb-1M and CASIA-Webface is used as the dataset. |
ULSee - Face Team | Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks https://arxiv.org/abs/1604.02878 Large-Margin Softmax Loss for Convolutional Neural Networks https://arxiv.org/abs/1612.02295 A Discriminative Deep Feature Learning Approach for Face Recognition https://ydwen.github.io/papers/WenECCV16.pdf NormFace: L2 Hypersphere Embedding for Face Verification https://arxiv.org/abs/1704.06369 SphereFace: Deep Hypersphere Embedding for Face Recognition https://arxiv.org/abs/1704.08063 ArcFace: Additive Angular Margin Loss for Deep Face Recognition https://arxiv.org/abs/1801.07698 |
SIATMMLAB TencentVision | Adopt the ensemble of very deep CNNs, learned by joint supervision (softmax loss, improved center loss, etc). Training data is a combination of public datasets (CAISA, VGG, CACD2000, etc) and private datasets. The total number of images is more than 2 million. |
DeepSense - Small | Adopt a network of very deep ResNet CNNs, learned by combined supervision(identification loss(softmax loss), verification loss, triplet loss). |
GRCCV | The algorithm consists of three parts: FCN - based fast face detection algorithm, pre-training ResNet CNN on classification task, weight tuning. Training set contains 273000 photos. Hardware: 8 x Tesla k80. |
SphereFace - Small | SphereFace uses a novel approach to learn face features that are discriminative on a hypersphere manifold. The training data set we use in SphereFace is the publicly available CASIA-WebFace dataset which contains 490k images of nearly 10,500 individuals. |
EM-DATA | arcface, https://github.com/deepinsight/insightface |
StartDT-AI | we only use a single model trained on a ResNet-28 network joined with cosine loss and triplet loss on MS-Celeb-1M(74K identities, 4M images), refer to DeepVisage: Making face recognition simple yet with powerful generalization skills https://arxiv.org/abs/1703.08388 One-shot Face Recognition by Promoting Underrepresented Classes https://arxiv.org/abs/1707.05574v2 ArcFace: Additive Angular Margin Loss for Deep Face Recognition https://arxiv.org/abs/1801.07698 SphereFace: Deep Hypersphere Embedding for Face Recognition https://arxiv.org/abs/1704.08063 |
BingMMLab-v1 (non-cleaned data) | Compares to our submissions named “BingMMLab V1(cleaned data)”, the only difference is that no data cleaning is adopted in this evaluation. |
BingMMLab V1(iBUG cleaned data) | We used knowledge graph to collect identities and then crawled Bing search engine to get high quality images, we filtered noises in 14M training data by clustering with weak face models, and then trained a DensetNet-69 (k=48) network with A-Softmax loss variants. In evaluation, we used the cleaned test set released by iBUG_DeepInsight. DenseNet https://arxiv.org/pdf/1608.06993v2.pdf A-Softmax and its variant: https://arxiv.org/pdf/1704.08063.pdf https://github.com/wy1iu/LargeMargin_Softmax_Loss/issues/13 |
- -   uses large training set
- -   uses large training set
- -   uses large training set
Algorithm | Date Submitted | Set 1 | Set 2 | Set 3 | Data Set Size |
---|---|---|---|---|---|
BingMMLab V1(iBUG cleaned data) | 4/10/2018 | 99.487% | 99.487% | 99.487% | Large |
iBUG_DeepInsight | 2/8/2018 | 98.948% | 98.948% | 99.008% | Large |
BingMMLab-v1 (non-cleaned data) | 4/10/2018 | 98.279% | 98.279% | 98.279% | Large |
TencentAILab_FaceCNN_v1 | 9/21/2017 | 97.961% | 97.961% | 97.961% | Large |
FaceTag V1 | 12/18/2017 | 97.163% | 97.163% | 97.163% | Large |
Argus_v1 | 2/6/2018 | 96.66% | 96.65% | 96.65% | Large |
EM-DATA | 4/4/2018 | 96.653% | 96.653% | 96.653% | Large |
SuningUS_AILab | 3/21/2018 | 96.19355202% | 96.19355202% | 96.19355202% | Large |
DeepSense V2 | 1/22/2017 | 95.993% | 95.993% | 94.984% | Large |
Yang Sun | 06/05/2017 | 95.178% | 96.179% | 96.187% | Large |
Vocord - deepVo V3 | 04/27/2017 | 94.963% | 94.963% | 94.963% | Large |
Video++ | 1/5/2018 | 94.89% | 94.89% | 94.89% | Large |
Intellivision | 2/11/2018 | 94.786% | 94.786% | 94.786% | Large |
Faceter Lab | 12/18/2017 | 94.749% | 94.749% | 94.749% | Large |
CVTE V2 | 1/27/2018 | 94.417% | 94.417% | 94.417% | Small |
StartDT-AI | 4/16/2018 | 93.8226% | 93.8226% | 93.8226% | Large |
XT-tech V2 | 08/30/2017 | 93.695% | 93.695% | 93.695% | Large |
THU CV-AI Lab | 12/12/2017 | 93.293% | 93.293% | 93.293% | Large |
ULSee - Face Team | 3/27/2018 | 93.235% | Large | ||
iBug (Reported by Author) | 04/28/2017 | 92.639% | Small | ||
YouTu Lab (Tencent Best-Image) | 04/08/2017 | 91.34% | 91.34% | 91.34% | Large |
SphereFace - Small | 12/1/2016 | 90.045% | 89.355% | 90.045% | Small |
TUPUTECH | 12/22/2017 | 88.726% | 88.726% | 88.726% | Large |
Progressor | 09/13/2017 | 88.59% | 85.59% | 85.59% | Large |
DeepSense - Large | 07/31/2016 | 87.764% | 87.764% | 87.764% | Large |
SIATMMLAB TencentVision | 12/1/2016 | 87.272% | 87.021% | 87.021% | Large |
Google - FaceNet v8 | 10/23/2015 | 86.473% | 86.386% | 86.473% | Large |
Shanghai Tech | 08/13/2016 | 86.369% | 86.369% | 86.369% | Large |
ForceInfo | 04/07/2017 | 85.918% | 85.619% | 85.619% | Large |
OceanAI Tech | 12/15/2017 | 85.86% | 85.86% | 85.891% | Large |
CVTE | 08/30/2017 | 85.563% | 85.563% | 85.563% | Large |
3DiVi Company - tdvm V2 | 04/15/2017 | 85.411% | 85.411% | 85.411% | Large |
NTechLAB - facenx_large | 10/20/2015 | 85.081% | 85.081% | 85.081% | Large |
DeepSense - Small | 07/31/2016 | 82.851% | 82.851% | 82.851% | Small |
Fudan University - FUDAN-CS_SDS | 1/29/2017 | 79.199% | 79.199% | 79.199% | Small |
Beijing Faceall Co. - FaceAll V2 | 04/28/2017 | 77.607% | 77.607% | 77.607% | Small |
Vocord-deepVo1.2 | 12/1/2016 | 77.143% | 77.143% | 77.143% | Large |
SIAT_MMLAB | 3/29/2016 | 76.72% | 76.72% | 76.72% | Small |
GRCCV | 12/1/2016 | 74.887% | 74.887% | 75.918% | Small |
Vocord - DeepVo1 | 08/03/2016 | 67.318% | 67.318% | 67.318% | Large |
Beijing Faceall Co. - FaceAll_Norm_1600 | 10/19/2015 | 67.118% | 67.118% | 67.118% | Large |
NTechLAB - facenx_small | 10/20/2015 | 66.366% | 66.427% | 66.427% | Small |
Beijing Faceall Co. - FaceAll_1600 | 10/19/2015 | 63.96% | 64.983% | 63.96% | Large |
Barebones_FR - cnn | 10/21/2015 | 59.036% | 59.036% | 59.036% | Small |
3DiVi Company - tdvm6 | 10/27/2015 | 36.927% | 37.967% | 36.927% | Small |
Joint Bayes | 10/20/2015 | 2.173% | 2.204% | 2.204% | Small |
LBP | 10/20/2015 | 1.465% | 1.465% | 1.465% | Small |
Algorithm | Details |
---|
- -   uses large training set
- -   uses large training set
- -   uses large training set
The colors represent identification accuracy going from 0(=blue)–none of the true pairs were matched to 1(=red)–all possible combinations of probe and gallery were matched per probe and gallery ages. White color indicates combinations of poses that did not exist in our test set.