Analyze outdoor walking data
Software prerequisites:
-
Keras
-
Tensorflow
-
numpy, scipy
-
h5pyi
-
sklearn, statsmodels
-
Setup experiment recording scripts
-
Condor scripts
- 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?
- 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
- 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