person identification => (1) FSA-Net을 통한 head pose estimation (2)VGG tranfer learning을 통한 Emotion recognition
대선토론 영상을 분석하여 의미 있는 데이터를 수집 Ex) 출연 비중, 후보자 간의 상호작용, 감정, 자세
# MTCNN face detector (slow but accurate)
cd demo
sh run_demo_FSANET_mtcnn.sh
=> Jy_demo_FSANET_mtcnn.py 실행
# SSD face detector (fast and accurate)
cd demo
sh run_demo_FSANET_ssd.sh
=> Jy_demo_FSANET.py 실행
[CVPR19] FSA-Net: Learning Fine-Grained Structure Aggregation for Head Pose Estimation from a Single Image Code Author: Tsun-Yi Yang
https://github.com/shamangary/FSA-Net/blob/master/0191.pdf
Tsun-Yi Yang, Yi-Ting Chen, Yen-Yu Lin, and Yung-Yu Chuang
This paper proposes a method for head pose estimation from a single image. Previous methods often predicts head poses through landmark or depth estimation and would require more computation than necessary. Our method is based on regression and feature aggregation. For having a compact model, we employ the soft stagewise regression scheme. Existing feature aggregation methods treat inputs as a bag of features and thus ignore their spatial relationship in a feature map. We propose to learn a fine-grained structure mapping for spatially grouping features before aggregation. The fine-grained structure provides part-based information and pooled values. By ultilizing learnable and non-learnable importance over the spatial location, different variant models as a complementary ensemble can be generated. Experiments show that out method outperforms the state-of-the-art methods including both the landmark-free ones and the ones based on landmark or depth estimation. Based on a single RGB frame as input, our method even outperforms methods utilizing multi-modality information (RGB-D, RGB-Time) on estimating the yaw angle. Furthermore, the memory overhead of the proposed model is 100× smaller than that of previous methods.
- Keras
- Tensorflow
- GTX-1080Ti
- Ubuntu
python 3.5.6 hc3d631a_0
keras-applications 1.0.4 py35_1 anaconda
keras-base 2.1.0 py35_0 anaconda
keras-gpu 2.1.0 0 anaconda
keras-preprocessing 1.0.2 py35_1 anaconda
tensorflow 1.10.0 mkl_py35heddcb22_0
tensorflow-base 1.10.0 mkl_py35h3c3e929_0
tensorflow-gpu 1.10.0 hf154084_0 anaconda
cudnn 7.1.3 cuda8.0_0
cuda80 1.0 0 soumith
numpy 1.15.2 py35_blas_openblashd3ea46f_0 [blas_openblas] conda-forge
numpy-base 1.14.3 py35h2b20989_0
- A guide for most dependencies. (in Chinese) http://shamangary.logdown.com/posts/3009851
- Anaconda
- OpenCV
- MTCNN
pip3 install mtcnn
- Capsule: https://github.com/XifengGuo/CapsNet-Keras
- Loupe_Keras: https://github.com/shamangary/LOUPE_Keras
There are three different section of this project.
- Data pre-processing
- Training and testing
- Demo
We will go through the details in the following sections.
This repository is for 300W-LP, AFLW2000, and BIWI datasets.
If you don't want to re-download every dataset images and do the pre-processing again, or maybe you don't even care about the data structure in the folder. Just download the file data.zip from the following link, and replace the data folder.
Now you can skip to the "Training and testing" stage.
In the paper, we define Protocol 1 and Protocol 2.
# Protocol 1
Training: 300W-LP (A set of subsets: {AFW.npz, AFW_Flip.npz, HELEN.npz, HELEN_Flip.npz, IBUG.npz, IBUG_Flip.npz, LFPW.npz, LFPW_Flip.npz})
Testing: AFLW2000.npz or BIWI_noTrack.npz
# Protocol 2
Training: BIWI(70%)-> BIWI_train.npz
Testing: BIWI(30%)-> BIWI_test.npz
(Note that type1 (300W-LP, AFLW2000) datasets have the same image arrangement, and I categorize them as type1. It is not about Protocal 1 or 2.)
If you want to do the pre-processing from the beginning, you need to download the dataset first.
Put 300W-LP and AFLW2000 folders under data/type1/, and put BIWI folder under data/
# For 300W-LP and AFLW2000 datasets
cd data/type1
sh run_created_db_type1.sh
# For BIWI dataset
cd data
python TYY_create_db_biwi.py
python TYY_create_db_biwi_70_30.py
# Training
sh run_fsanet_train.sh
# Testing
# Note that we calculate the MAE of yaw, pitch, roll independently, and average them into one single MAE for evaluation.
sh run_fsanet_test.sh
Just remember to check which model type you want to use in the shell script and you are good to go.
You need a webcam to correctly process the demo file.
Note the the center of the color axes is the detected face center. Ideally, each frame should have new face detection results. However, if the face detection fails, the previous detection results will be used to estimate poses.
LBP is fast enough for real-time face detection, while MTCNN is much more accurate but slow.
(2019/08/30 update!) SSD face detection is robust and fast! I borrow some face detector code from https://www.pyimagesearch.com
# LBP face detector (fast but often miss detecting faces)
cd demo
sh run_demo_FSANET.sh
# MTCNN face detector (slow but accurate)
cd demo
sh run_demo_FSANET_mtcnn.sh
# SSD face detector (fast and accurate)
cd demo
sh run_demo_FSANET_ssd.sh
cd training_and_testing
python keras_to_tf.py --trained-model-dir-path ../pre-trained/300W_LP_models/fsanet_var_capsule_3_16_2_21_5 --output-dir-path <your_output_dir>Above command will generate the tensorflow frozen graph in <your_output_dir>/converted-models/tf/fsanet_var_capsule_3_16_2_21_5.pb
- ssr_G_model:
https://github.com/shamangary/FSA-Net/blob/master/lib/FSANET_model.py#L441
- Two-stream structure for extracting the features.
- ssr_feat_S_model:
https://github.com/shamangary/FSA-Net/blob/master/lib/FSANET_model.py#L442
- Generating fine-grained structure mapping from different scoring functions.
- Apply the mapping on to the features and generate primary capsules.
- ssr_aggregation_model:
https://github.com/shamangary/FSA-Net/blob/master/lib/FSANET_model.py#L443
- Feed the primary capsules into capsule layer and output the final aggregated capsule features. And divide them into 3 stages.
- ssr_F_model:
https://github.com/shamangary/FSA-Net/blob/master/lib/FSANET_model.py#L444
- Taking the previous 3 stages features for Soft-Stagewise Regression (SSR) module. Each stage further splits into three parts: prediction, dynamic index shifting, and dynamic scaling. This part please check the '[IJCAI18] SSR-Net' for more detail explanation.
- SSRLayer:
https://github.com/shamangary/FSA-Net/blob/master/lib/FSANET_model.py#L444
- Taking the prediction, dynamic index shifting, and dynamic scaling for the final regression output. In this case, there are three outputs (yaw, pitch, roll).
Awesome VR flight simluation with FSANet ONNX format