際際滷

際際滷Share a Scribd company logo

Face recognition with Deep Neural Network

Erkhembaatar M.
Erkhembaatar M.

Face recognition with Convolutional Neural Network

1 of 18
Downloaded 31 times
亟仍亳亶仆 亞勵仆 勵仍亢 舒亳亞仍舒仆 勵仆亳亶 舒舒亶 舒仆亳 舒亞舒仍舒仍仆 亟舒仍亞舒舒 (Face Recognition with Deep Neural Network) .亅仄弍舒舒舒 .丱勵亟* .弍舒舒舒** 舒亞亳舒仆, 仂仄仗ﾑ亠亳亶仆 丕舒舒仆 舒仍弍舒, 丿丕丐弌-丱丐弌, 丕仍舒舒仆弍舒舒舒, 仂仆亞仂仍 仍 *丕亟亳亟舒亞: 仂从仂, 亟亟 仗仂., 仂仄仗ﾑ亠亳亶仆 丕舒舒仆 舒仍弍舒, 丿丕丐弌-丱丐弌, 丕仍舒舒仆弍舒舒舒, 仂仆亞仂仍 仍 **唏于仍唏: 仂从仂舒仆 舒仍舒 弍舒亞, 亅仍亠从仂仆亳从亳亶仆 舒仍弍舒, 丿丕丐弌-丱丐弌, 丕仍舒舒仆弍舒舒舒, 仂仆亞仂仍 仍 2016-11-30
丕亟亳亞舒仍 勵 弍仂仍仂仆 舒舒亶 亳仍勵勵仍, 舒仆亳 亞 唏唏仆 丱勵仆亳亶 舒亟舒于亳 99.5%, DeepFace 99.7%, EigenFace 64.8% 舒 亳 仄亢仆亳亶 唏亞唏亞亟唏仍: ImageNet 舒舒仍仍亠仍 仂仂仂仂仍仂仍: GPU, CUDA, cuDNN, Caffe, Torch 丐舒舒仍亞舒舒仆 从舒仄亠舒舒 亟勵, 舒舒亶 舒仆亳 亞 丱勵仆亟仍, 仂仂仂仂仍仍仆 亰舒亟舒仍 亞亳亶亞 弌仆仂亟仆 亳 亞仍亳亶仆 CS131: Computer Vision: Foundations and Applications 亳仍亳亶仆 仄舒亠亳舒仍亳舒 舒于舒于 亰亞亳亶亞 Brandon Amos
丱亞: 仂仄仗ﾑ亠 舒舒舒 丱亳亶仄仍 仂ﾑ仆 舒舒仆, 仂弍仂 丕舒舒仍舒亞 亞, 仂亳, 仂 亅勵勵仍 仄仆亟, 仂仆仂亳仍亞仂仂 CT, MRI (丱仂 舒于亟舒 亳仍勵勵仍) 仂仆, 亢仂仍仂仂亞勵亶 舒于仂仄舒亳仆 舒仄亠亳亶仆 礌舒仍, 亞舒亰舒亰勵亶 仂仆 亰舒亞仍舒仍 亳亰仆亠, 仄亠亟亳舒, 仆亠舒亶仆仄亠仆, ... 亞亟亞 Nvidia.com, Omate.com 舒亶舒舒 舒于舒于
仂亳仍亞仂 仆仂仍 舒仍亞仂亳仄亟 舒仍舒亢, 亰仄亳 Computer Vision, Image Classification, Machine & Deep Learning, CNN, RNN, Softmax, SVM, .. 哦亞唏亞亟仍亳亶仆 弍舒舒亰 弍勵亟勵勵仍 哦亞唏亞亟唏仍 仂仍弍仂仍仂仍, 弌亞舒仍仆 唏亞唏亞亟唏仍, 唏亞唏亞亟唏仍 (ImageNet 14 舒 亰舒亞舒亶) ... 丐亠仆亳从 舒亟于舒 亰仄亳 弌仗亠从仂仄仗ﾑ亠: GPU, 舒舒仍仍亠仍 仂仂仂仂仍仂仍, .. 舒亞舒亢 亞仍: Numpy, Scikit-learn, Linux, OpenCV, Dlib, CUDA, cuDNN, Caffe, Torch, TensorFlow, .. 丐亳仍, 亟舒仍亞舒舒仆 舒亞舒仍舒仍仆 舒亟于舒 亰仄亳 勵 亳仍勵勵仍, 舒舒亶 舒仆亳, 舒仆亞亳仍舒亞, 仄亟仍亳亶仆 勵仍亢 勵勵亞 亳仆 舒弍仂舒仂亳, 丱丐弌 NVIDIA GPU Educators Program, AI & Autonomous Robotic 亳仍, 仍舒弍. 舒亢亳仍 AI Robot of smart home
弌亟仍舒亞亟舒仆 弍舒亶亟舒仍 Pre-trained VGG models by Oxford University, 仂仆亞仂仍仆 亢亳*
丶舒舒亶 亳仍勵勵仍, 舒仆亳 (亰亞亳亶亞 Face Detection and Recognition: Theory and Practice 仆仂仄仆仂仂 舒于 舒亳亞仍舒于)
丶舒舒亶 亳仍勵勵仍: HOG 于 亳仆亢 Histogram of Oriented Gradients 1. 丶亞亳亶仆 亞舒亟亳亠仆 亳亞仍仍 2. HOG 亟勵仍仍 3. HOG 舒舒亶仆 于 亳仆亢 亰亞亳亶亞 Machine Learning is Fun! Part 4: Modern Face Recognition with Deep Learning 仆亳亶仍仍 舒于 舒亳亞仍舒于
亟仍亳亶仆 亞勵仆 勵仍亢 Convolutional, Non-linear, Pooling (Down sampling), Fully Connected layers, Output 亞亳亶亞 A Beginner's Guide To Understanding Convolutional Neural Networks 仆亳亶仍仍 舒于 舒亳亞仍舒于
CNN: 丿勵勵仍勵勵亟 (Convolving) 亞亳亶亞 A Beginner's Guide To Understanding Convolutional Neural Networks 仆亳亶仍仍 舒于 舒亳亞仍舒于
CNN: 亟仍 (neuron) 亞亳亶亞 A Beginner's Guide To Understanding Convolutional Neural Networks 仆亳亶仍仍 舒于 舒亳亞仍舒于
CNN: 丿勵勵仍勵勵 (feature) 亞亳亶亞 A Beginner's Guide To Understanding Convolutional Neural Networks 仆亳亶仍仍 舒于 舒亳亞仍舒于 亰亞亳亶亞 Andrew Ng
CNN: Backpropagation Forward pass, Lost function, Backward pass, Weight update MSE (Mean Squared Error), Softmax, SVM, Gradient descent, Epoch 亞亳亶亞 A Beginner's Guide To Understanding Convolutional Neural Networks 仆亳亶仍仍 舒于 舒亳亞仍舒于
丶舒舒亶 舒仆亳仍: 弌亞舒仍仆 唏亞唏亞亟唏仍 4 勵仆亳亶 仆亳亶 226 亞舒仍仆 唏亞唏亞亟唏仍, 亳仍仆 50 亰舒亞 Dlib, OpenCV, Python
丶舒舒亶 舒仆亳仍: 丱亞亢勵勵仍仍, 亳仍 Python, OpenCV, Dlib, Torch, OpenFace, CUDA, cuDNN, LinuxMint HOG face detection, pre-trained CNN model with Linear SVM classifier
勳 亟勵仆 丐亳仍仆 亟仆亟舒亢 92.54%, DeepFace 99.97%, 丱勵仆亳亶 舒亟舒于亳 99.5% 哦亞唏亞亟唏仍 丐亳仍仆 勵 亟勵仆 (True positive) 仆亟舒亢 仍亟舒舒 丱勵勵亟, 2 仆舒 50 0.83 0.87 0.91 0.98 0.86 0.97 0.97 0.99 0.97 0.92 0.927 0 丱勵勵亟, 8 仆舒 68 0.99 0.99 0.68 0.97 0.99 0.94 1.00 1.00 0.99 0.48 0.903 2 丐仂仄 勵仆, 26 0.92 0.95 0.90 0.78 0.90 0.93 0.87 0.97 0.96 0.93 0.911 1 丐仂仄 勵仆, 仄 82 0.83 0.98 0.99 0.91 0.93 0.98 0.99 0.98 1.00 0.99 0.958 0 亳亶 226 0.925 3 哦亞唏亞亟唏仍 丐亳仍仆 勵 亟勵仆 (True negative) 仆亟舒亢 仍亟舒舒 哦唏 勵仆 10 0.49 0.59 0.42 0.21 0.28 0.53 0.03 0.06 0.64 0.54 0.379 6
勵亞仆仍 仆亳仆亳从 亅从亳仆亳从 亢亳亞仍舒亞舒舒 舒仄舒舒舒仍亞勵亶 亢亳亞仍舒亞舒舒 舒仄舒舒舒仍舒亶 个亳亰亳从 勵亳仆 亰勵亶仍 仍勵勵仍亞, 勳仆亳亶 仍弍, 仆舒亢亳仍 舒亶亳仍 勵勵仆亳亶 于亳舒仍, 仆勵亟仆亳亶 亳仍 个仂从 仆 唏仆亞唏, 勵亶, 亞舒舒舒仆亰勵亶仆 丿亞亳舒仆 (noise) 弌亞舒仍仆 唏亞唏亞亟唏仍 唏唏亟唏仆, 弍舒亟 勵仄勵勵亳亶仆 亰舒亞 仂仍舒 92.5% 仄亢仆亳亶 唏亞唏亞亟唏仍 (勵仆 弍勵亳亶仆 500-1000) 亟 GPU 舒亳亞仍舒亢 亳仍 亳亶 丐舒舒仍亞舒舒仆 从舒仄亠, 亞亳亶仆 仍舒-仂弍仂仆 亳亶仄仍 仂ﾑ仆亟 舒亳亞仍舒 弍仂仍仂仄亢仂亶
亳亞仍舒舒仆 仄舒亠亳舒仍 [1] Baidus Artificial-Intelligence Supercomputer Beats Google at Image Recognition, MIT Technology Review, 2015 [2] DeepFace: Closing the Gap to Human-Level Performance in Face Verification. Facebook AI Research Publication, 2014 [3] Machine Learning is Fun! Part 4: Modern Face Recognition with Deep Learning, 2016 [4] Navneet Dalal, Bill Triggs. "Histograms of Oriented Gradients for Human Detection, 2005 [5] Vahid Kazemi, Josephine Sullivan. One Millisecond Face Alignment with an Ensemble of Regression Trees, 2014 [6] Florian Schroff, Dmitry Kalenichenko, James Philbin. FaceNet: A Unified Embedding for Face Recognition and Clustering, 2015 [7] Brandon Amos. OpenFace. https://cmusatyalab.github.io/openface/ [8] D. A. Forsyth and J. Ponce. "Computer Vision: A Modern Approach (2nd edition)". Prence Hall, 2011 [9] opencv.org, dlib.com, http://torch.ch [10] CUDA, cuDNN. http://nvidia.com [11] CS231n Convolutional Neural Networks for Visual Recognition, Stanford University [12] Stan Z. Li Anil K. Jain. Handbook of Face Recognition. Springer, 2004 [13] Asit Kumar Datta, Madhura Datta, Pradipta Kumar Banerjee. Face Detection and Recognition: Theory and Practice. Taylor & Francis, 2015 [14] Mohamed Daoudi, Anuj Srivastava, Remco Veltkamp. 3D Face Modeling, Analysis and Recognition. Wiley, 2013
仆舒舒舒仍 舒仆亟仍舒仆亟 弍舒仍舒仍舒舒! 仍? Engineers turn dreams into reality Hayao Miyazaki

More Related Content

Face recognition with Deep Neural Network

  • 1. 亟仍亳亶仆 亞勵仆 勵仍亢 舒亳亞仍舒仆 勵仆亳亶 舒舒亶 舒仆亳 舒亞舒仍舒仍仆 亟舒仍亞舒舒 (Face Recognition with Deep Neural Network) .亅仄弍舒舒舒 .丱勵亟* .弍舒舒舒** 舒亞亳舒仆, 仂仄仗ﾑ亠亳亶仆 丕舒舒仆 舒仍弍舒, 丿丕丐弌-丱丐弌, 丕仍舒舒仆弍舒舒舒, 仂仆亞仂仍 仍 *丕亟亳亟舒亞: 仂从仂, 亟亟 仗仂., 仂仄仗ﾑ亠亳亶仆 丕舒舒仆 舒仍弍舒, 丿丕丐弌-丱丐弌, 丕仍舒舒仆弍舒舒舒, 仂仆亞仂仍 仍 **唏于仍唏: 仂从仂舒仆 舒仍舒 弍舒亞, 亅仍亠从仂仆亳从亳亶仆 舒仍弍舒, 丿丕丐弌-丱丐弌, 丕仍舒舒仆弍舒舒舒, 仂仆亞仂仍 仍 2016-11-30
  • 2. 丕亟亳亞舒仍 勵 弍仂仍仂仆 舒舒亶 亳仍勵勵仍, 舒仆亳 亞 唏唏仆 丱勵仆亳亶 舒亟舒于亳 99.5%, DeepFace 99.7%, EigenFace 64.8% 舒 亳 仄亢仆亳亶 唏亞唏亞亟唏仍: ImageNet 舒舒仍仍亠仍 仂仂仂仂仍仂仍: GPU, CUDA, cuDNN, Caffe, Torch 丐舒舒仍亞舒舒仆 从舒仄亠舒舒 亟勵, 舒舒亶 舒仆亳 亞 丱勵仆亟仍, 仂仂仂仂仍仍仆 亰舒亟舒仍 亞亳亶亞 弌仆仂亟仆 亳 亞仍亳亶仆 CS131: Computer Vision: Foundations and Applications 亳仍亳亶仆 仄舒亠亳舒仍亳舒 舒于舒于 亰亞亳亶亞 Brandon Amos
  • 3. 丱亞: 仂仄仗ﾑ亠 舒舒舒 丱亳亶仄仍 仂ﾑ仆 舒舒仆, 仂弍仂 丕舒舒仍舒亞 亞, 仂亳, 仂 亅勵勵仍 仄仆亟, 仂仆仂亳仍亞仂仂 CT, MRI (丱仂 舒于亟舒 亳仍勵勵仍) 仂仆, 亢仂仍仂仂亞勵亶 舒于仂仄舒亳仆 舒仄亠亳亶仆 礌舒仍, 亞舒亰舒亰勵亶 仂仆 亰舒亞仍舒仍 亳亰仆亠, 仄亠亟亳舒, 仆亠舒亶仆仄亠仆, ... 亞亟亞 Nvidia.com, Omate.com 舒亶舒舒 舒于舒于
  • 4. 仂亳仍亞仂 仆仂仍 舒仍亞仂亳仄亟 舒仍舒亢, 亰仄亳 Computer Vision, Image Classification, Machine & Deep Learning, CNN, RNN, Softmax, SVM, .. 哦亞唏亞亟仍亳亶仆 弍舒舒亰 弍勵亟勵勵仍 哦亞唏亞亟唏仍 仂仍弍仂仍仂仍, 弌亞舒仍仆 唏亞唏亞亟唏仍, 唏亞唏亞亟唏仍 (ImageNet 14 舒 亰舒亞舒亶) ... 丐亠仆亳从 舒亟于舒 亰仄亳 弌仗亠从仂仄仗ﾑ亠: GPU, 舒舒仍仍亠仍 仂仂仂仂仍仂仍, .. 舒亞舒亢 亞仍: Numpy, Scikit-learn, Linux, OpenCV, Dlib, CUDA, cuDNN, Caffe, Torch, TensorFlow, .. 丐亳仍, 亟舒仍亞舒舒仆 舒亞舒仍舒仍仆 舒亟于舒 亰仄亳 勵 亳仍勵勵仍, 舒舒亶 舒仆亳, 舒仆亞亳仍舒亞, 仄亟仍亳亶仆 勵仍亢 勵勵亞 亳仆 舒弍仂舒仂亳, 丱丐弌 NVIDIA GPU Educators Program, AI & Autonomous Robotic 亳仍, 仍舒弍. 舒亢亳仍 AI Robot of smart home
  • 5. 弌亟仍舒亞亟舒仆 弍舒亶亟舒仍 Pre-trained VGG models by Oxford University, 仂仆亞仂仍仆 亢亳*
  • 6. 丶舒舒亶 亳仍勵勵仍, 舒仆亳 (亰亞亳亶亞 Face Detection and Recognition: Theory and Practice 仆仂仄仆仂仂 舒于 舒亳亞仍舒于)
  • 7. 丶舒舒亶 亳仍勵勵仍: HOG 于 亳仆亢 Histogram of Oriented Gradients 1. 丶亞亳亶仆 亞舒亟亳亠仆 亳亞仍仍 2. HOG 亟勵仍仍 3. HOG 舒舒亶仆 于 亳仆亢 亰亞亳亶亞 Machine Learning is Fun! Part 4: Modern Face Recognition with Deep Learning 仆亳亶仍仍 舒于 舒亳亞仍舒于
  • 8. 亟仍亳亶仆 亞勵仆 勵仍亢 Convolutional, Non-linear, Pooling (Down sampling), Fully Connected layers, Output 亞亳亶亞 A Beginner's Guide To Understanding Convolutional Neural Networks 仆亳亶仍仍 舒于 舒亳亞仍舒于
  • 9. CNN: 丿勵勵仍勵勵亟 (Convolving) 亞亳亶亞 A Beginner's Guide To Understanding Convolutional Neural Networks 仆亳亶仍仍 舒于 舒亳亞仍舒于
  • 10. CNN: 亟仍 (neuron) 亞亳亶亞 A Beginner's Guide To Understanding Convolutional Neural Networks 仆亳亶仍仍 舒于 舒亳亞仍舒于
  • 11. CNN: 丿勵勵仍勵勵 (feature) 亞亳亶亞 A Beginner's Guide To Understanding Convolutional Neural Networks 仆亳亶仍仍 舒于 舒亳亞仍舒于 亰亞亳亶亞 Andrew Ng
  • 12. CNN: Backpropagation Forward pass, Lost function, Backward pass, Weight update MSE (Mean Squared Error), Softmax, SVM, Gradient descent, Epoch 亞亳亶亞 A Beginner's Guide To Understanding Convolutional Neural Networks 仆亳亶仍仍 舒于 舒亳亞仍舒于
  • 13. 丶舒舒亶 舒仆亳仍: 弌亞舒仍仆 唏亞唏亞亟唏仍 4 勵仆亳亶 仆亳亶 226 亞舒仍仆 唏亞唏亞亟唏仍, 亳仍仆 50 亰舒亞 Dlib, OpenCV, Python
  • 14. 丶舒舒亶 舒仆亳仍: 丱亞亢勵勵仍仍, 亳仍 Python, OpenCV, Dlib, Torch, OpenFace, CUDA, cuDNN, LinuxMint HOG face detection, pre-trained CNN model with Linear SVM classifier
  • 15. 勳 亟勵仆 丐亳仍仆 亟仆亟舒亢 92.54%, DeepFace 99.97%, 丱勵仆亳亶 舒亟舒于亳 99.5% 哦亞唏亞亟唏仍 丐亳仍仆 勵 亟勵仆 (True positive) 仆亟舒亢 仍亟舒舒 丱勵勵亟, 2 仆舒 50 0.83 0.87 0.91 0.98 0.86 0.97 0.97 0.99 0.97 0.92 0.927 0 丱勵勵亟, 8 仆舒 68 0.99 0.99 0.68 0.97 0.99 0.94 1.00 1.00 0.99 0.48 0.903 2 丐仂仄 勵仆, 26 0.92 0.95 0.90 0.78 0.90 0.93 0.87 0.97 0.96 0.93 0.911 1 丐仂仄 勵仆, 仄 82 0.83 0.98 0.99 0.91 0.93 0.98 0.99 0.98 1.00 0.99 0.958 0 亳亶 226 0.925 3 哦亞唏亞亟唏仍 丐亳仍仆 勵 亟勵仆 (True negative) 仆亟舒亢 仍亟舒舒 哦唏 勵仆 10 0.49 0.59 0.42 0.21 0.28 0.53 0.03 0.06 0.64 0.54 0.379 6
  • 16. 勵亞仆仍 仆亳仆亳从 亅从亳仆亳从 亢亳亞仍舒亞舒舒 舒仄舒舒舒仍亞勵亶 亢亳亞仍舒亞舒舒 舒仄舒舒舒仍舒亶 个亳亰亳从 勵亳仆 亰勵亶仍 仍勵勵仍亞, 勳仆亳亶 仍弍, 仆舒亢亳仍 舒亶亳仍 勵勵仆亳亶 于亳舒仍, 仆勵亟仆亳亶 亳仍 个仂从 仆 唏仆亞唏, 勵亶, 亞舒舒舒仆亰勵亶仆 丿亞亳舒仆 (noise) 弌亞舒仍仆 唏亞唏亞亟唏仍 唏唏亟唏仆, 弍舒亟 勵仄勵勵亳亶仆 亰舒亞 仂仍舒 92.5% 仄亢仆亳亶 唏亞唏亞亟唏仍 (勵仆 弍勵亳亶仆 500-1000) 亟 GPU 舒亳亞仍舒亢 亳仍 亳亶 丐舒舒仍亞舒舒仆 从舒仄亠, 亞亳亶仆 仍舒-仂弍仂仆 亳亶仄仍 仂ﾑ仆亟 舒亳亞仍舒 弍仂仍仂仄亢仂亶
  • 17. 亳亞仍舒舒仆 仄舒亠亳舒仍 [1] Baidus Artificial-Intelligence Supercomputer Beats Google at Image Recognition, MIT Technology Review, 2015 [2] DeepFace: Closing the Gap to Human-Level Performance in Face Verification. Facebook AI Research Publication, 2014 [3] Machine Learning is Fun! Part 4: Modern Face Recognition with Deep Learning, 2016 [4] Navneet Dalal, Bill Triggs. "Histograms of Oriented Gradients for Human Detection, 2005 [5] Vahid Kazemi, Josephine Sullivan. One Millisecond Face Alignment with an Ensemble of Regression Trees, 2014 [6] Florian Schroff, Dmitry Kalenichenko, James Philbin. FaceNet: A Unified Embedding for Face Recognition and Clustering, 2015 [7] Brandon Amos. OpenFace. https://cmusatyalab.github.io/openface/ [8] D. A. Forsyth and J. Ponce. "Computer Vision: A Modern Approach (2nd edition)". Prence Hall, 2011 [9] opencv.org, dlib.com, http://torch.ch [10] CUDA, cuDNN. http://nvidia.com [11] CS231n Convolutional Neural Networks for Visual Recognition, Stanford University [12] Stan Z. Li Anil K. Jain. Handbook of Face Recognition. Springer, 2004 [13] Asit Kumar Datta, Madhura Datta, Pradipta Kumar Banerjee. Face Detection and Recognition: Theory and Practice. Taylor & Francis, 2015 [14] Mohamed Daoudi, Anuj Srivastava, Remco Veltkamp. 3D Face Modeling, Analysis and Recognition. Wiley, 2013