Advanced Face Recognition Research Update
Stay updated on the latest advancements in face recognition technology with Adri Priadana's weekly reports. Explore activities such as paper preparations, journal submissions, and architecture modifications for enhanced accuracy and efficiency in face recognition models. Follow along as the team delves into ResNet architectures and continues to push the boundaries in the field of AI face recognition.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
Presentation Transcript
Weekly Report Adri Priadana January 31, 2023
Last Week Activities Review IW-FCV 2023 papers Continue to prepare for ISIE 2023 paper Face-based Age Group Recognizer Write the paper (Finished) Double check with Putro and Linh Plagiarism checker Report and Discuss with Professor via KakaoTalk 2
TII Journal Continue to try to improve or modify the ResNet architecture by developing some mechanism for Face Recognition (Target much more efficient with competitive or outperform accuracy) Try to train on decreased Dataset (24 epoch) Decrease the instance (Ins15) : 1,250,590 (1 epoch 1 hour 10 minutes) Result Model Parameters LFW CFP-FP AgeDB CALFW CPLFW Average R-50 44 M 99.67 96.34 97.22 95.88 91.40 96.10 R-50 (SE | GC+CBAM) 26 M 99.68 95.99 96.65 95.87 90.87 95.81 | : stage 1,2,3 | 4 of ResNet50 GC : Group Convolution SE : Squeeze and Excitation CBAM : Convolutional Block Attention module 3
This Week Activities ISIE 2023 paper Face-based Age Group Recognizer Double check and submit the paper Continue to write the Research Note TII Journal Continue to try to improve or modify the ResNet architecture by developing some mechanism for Face Recognition (Target much more efficient with competitive or outperform accuracy) Try to code and implement Pyramid Squeeze Attention (PSA) 4
ResNet 3, 4, 14, 3 6
Prepare for ISIE 2023 Conference Paper Performance Evaluation on UTKFace To Recognize Exact Age Performance Evaluation on UTKFace To Recognize Age Group SOTA Baseline Number of Parameters MAE Architectures Number of Parameters Validation Accuracy (%) OR-CNN (2016) Manually-designed ResNet34 MobileNetV1 ResNet34 ResNet50 ResNet50 Manually-designed VGG16 Manually-designed - 5.76 5.47 5.44 5.14 4.55 4.47 4.44 4.37 4.38 ResNet34 ResNet50 SquezeeNet + BN MobileNetV2 VGG11 + BN 2PDG (2022) VGG16 + BN VGG13 + BN Proposed 21,1 M 23.6 M 0.74 M 2.26 M 34.4 M 0.46 M 39.8 M 34.5 M 0,49 M 87.13 88.43 88.47 90.08 90.12 90.12 90.34 90.42 90.90 CORAL (2020) Savchenko (2019) LRTI (2022) Berg et al. (2020) FCRN (2022) 2PDG (2022) MWR (2022) Proposed 21 M 3.5 M 21 M 24 M 24 M 0.46 M 40 M 0.49 M M2L Residual on M2L DELA Number of Parameters 463,268 Validation Accuracy (%) Ablation Study on UTKFace To Recognize Age Group 90.08 90.21 90.25 90.90 482,020 482,020 486,822 10
Prepare for ISIE 2023 Conference Paper Performance Evaluation on FGNET To Recognize Exact Age SOTA Baseline Number of Parameters MAE LSDML (2018) ARAN (2019) M-LSDML (2018) DLDLF (2021) DRF (2021) DAG-VGG16 (2019) DAG-GoogleNet (2019) ADPF (2022) 2PDG (2022) MSFCL (2020) BridgeNet (2019) MWR (2022) Proposed ResNet101 VGG16 ResNet101 VGG16 VGG16 VGG16 GoogLeNet Manually-designed Manually-designed Manually-designed Manually-designed VGG16 Manually-designed 44 M 414 M 44 M 14 M 14 M 24 M 131 M 14 M 0.46 M 15 M 120 M 40 M 0.49 M 3.92 3.79 3.74 3.71 3.41 3.08 3.05 2.86 2.75 2.71 2.56 2.23 2.71 11
Prepare for ISIE 2023 Conference Paper Runtime Efficiency Architectures VGG16 VGG13 Parameter 39,790,916 34,476,100 GFLOPs 2.2900 1.6100 Age (FPS) 42.20 50.97 Face + Age (FPS) 36.25 42.65 VGG11 34,421,892 1.2700 55.12 45.37 ResNet50 ResNet34 MobileNetV2 SquezeeNet + BN 2PDG (2022) Proposed 23,595,908 21,105,284 2,263,108 736,340 459,476 486,822 0.6330 0.2170 0.0501 0.0833 0.0645 0.0406 56.02 80.70 121.00 230.67 316.23 330.66 46.02 61.28 81.79 122.19 144.00 144.49 12
Prepare for ISIE 2023 Conference Paper Comparisons of Different Attention Modules applied on the Proposed Architecture on UTKFace dataset with Setting II. Attention Modules applied in the Proposed Architecture BAM SE Number of Parameters Validation Accuracy GFLOPs Age (FPS) Face + Age (FPS) 489,481 482,532 0.0412 0.0405 320.14 342.64 141.92 147.29 90.51 90.55 CBAM 482,618 0.0405 335.86 145.76 90.77 Proposed 486,822 0.0406 330.66 144.49 90.90 13
TII Journal Continue to try to improve or modify the ResNet architecture by developing some mechanism for Face Recognition (Target much more efficient with competitive or outperform accuracy) Try to train on decreased Dataset (10 epoch) Decrease the instance (Ins15) : 1,250,590 (1 epoch 1 hour 10 minutes) Decrease the class (C20000) : 1,530,886 (1 epoch 1 hour 20 minutes) Result Model Parameters LFW CFP-FP AgeDB CALFW CPLFW Average R-50 on C20000 44 M 99.58 93.64 96.05 95.23 88.62 94.62 R-50 on Ins15 44 M 99.58 94.16 95.93 95.22 89.83 94.94 R-50 (SE | GC+CBAM) on Ins15 26 M 99.48 94.24 95.63 95.07 89.73 94.83 | : stage 1,2,3 | 4 of ResNet50 GC : Group Convolution SE : Squeeze and Excitation CBAM : Convolutional Block Attention module 14
