/3d_conv

Primary LanguagePython

3d_conv

Calendar

April 14 - Final presentation due

3d Convolution for Pylearn2

Project Overview

####contact_and_potential_grasps_small.h5 A small dataset containing 8 grasps. Each grasp has an rgbd image, and 4 uvd tuples. The uvd values are the pixel location and depth values for the palm, and three fingertips.

####rgbd_hdf5_dataset.py Wraps around an hdf5 datafile, and provides pairs of patches and labels at training time.

####train_model.py Call this to train a model. It allows you to specify a model.yaml file as well as a dataset to train with. The model is selected out of the model_templates directory and copied into the models directory.

####model_templates directory Contains .yaml files specifying layer configurations, learning rates, and other parameters needed to train a model

####models directory When a model is trained, all yaml files from the model_templates directory are copied over here. The .pkl file for the model is also stored here, along with images of the weights generated as the model trains.

####set_paths.sh Places this projects code on your PYTHONPATH and also sets HOME_PATH which is used in paths.py to determine where different directories are located.

####example usage

$ source set_paths.sh
$ python train_model.py