/Walking-Data-Analysis

Analyze outdoor walking data

Primary LanguagePython

Walking-Data-Analysis

Analyze outdoor walking data

Software prerequisites:

  • Keras

  • Tensorflow

  • numpy, scipy

  • h5pyi

  • sklearn, statsmodels

  • Setup experiment recording scripts

  • Condor scripts

Gaze prediction task 1: classification of image patches

  • Get gaze vs. nongaze data
  • Basic CNN classification model
  • Data augmentation -- preprocess, reduce overfitting
  • Colored data
  • Max pooling?
  • Optical flow only
  • Colored data + optical flow
  • 32x32 data
  • Convolution dilation rate
  • Fourier transform
  • Fully connected network
  • GlobalAveragePooling2D?
  • Load pre-trained network?

Gaze prediction task2: body-joint model

  • Body joint prediction: regression model in sklearn
  • Regression model interpretation: regression model in statsmodel
  • Variable names
  • Multicolinearity issue
  • Time series issue
  • Statsmodel regression with multi-dimensional output
  • MLP model
  • Sequence model

Gaze prediction task3: joint image-body model

  • Joint network
  • How to do this:
    • from body-model-1condition.py one can see how to read in body joints from file, and how to do regression using a fully connected network
    • from cnn-img3ch+opf.py one can see how to build a convolutional network with multiple input sources and fuse them
    • from these two above should be able to build a joint network that takes in both image and joints, and predict gaze