Here are some instrctions for training the network.

1. Tensorflow 1.3 or above.
2. opencv3

update soon....

update soon....

After pre-processing, you will have following files.

mean_pixelrgb.npy, std_pixelrgb.npy : the mean and the standard deviation of each pixel on images in the entire dataset.

training set:

sf_filename.npy : A list of colorized depth image path. (shuffle)

sf_vector.npy : A list of grasp parameters. (shuffle)

sf_label.npy : A list of label. (shuffle)

validation set

sf_vfilename.npy : A list of colorized depth image path. (shuffle)

sf_vvector.npy : A list of grasp parameters. (shuffle)

sf_vlabel.npy : A list of label. (shuffle)

Above files are the input of the network. Please see example floder to see the right format of above files.

It`s easy to train the network if you have above files. Just run following command:

python train.py

It will read above files and train the network for 40 epochs. The hyperparameters are easily to change in the train.py

Update soon.........