Deep learning for visual understanding

speaker: Jiashi Feng event: SG Self-driving Car Engineering Meetup

deep learning neural network

• stack of filters

convolution neural network

• convolution layer
• sub-sampling layer
• deep face system on facebook
• face detection and pose estimation
• scene parsing
• image/video superresolution
• implemented on caffe/tensor flow

occlusion-robust face alignment

• landmark detection

face de-occlusion

• joint detection and recovery

human re-identification

• Long short-term memory (LSTM)
• recurrent neural network
• implemented on theano

scene understanding

hyper parameters are tuned manually