Author: Akhil
2/09/19: Meeting Minutes:
- Progress report
- Akhil,Rintaro:Face Recognition Baseline: Preprocessing Dataset, improving inference speed of RetinaFace
- Pradeep: Research on GAN-Angle and MixUp
- Kaushik: Multitask learning: Coding end-to-end face recognition network.
- Sodoo: Embedding result based on different augmentation(darkening, brightening, blurring, angle)
- Sajjad: Classification result based on different augmentation(blurring, resizing, super resolution, super resolution with blur)
- Next step
- Create an entire flow till final recognition model as blocks
- Initially use the blocks as open source codes and calculate accuracies
- The blocks will be replaced one by one with our trained model and compared with the initial accuracies
- Check the final result with exawizard data
- Make it closer to real life by testing using iPad and edge
- Cross-validation
- Schedule:
- sept:- Create a proper flow with dlib models
- Oct:- Improve accuracy without considering speed of inference
- Nov:- speedup model and implement on edge