simonwsw
commented
7 years ago The four channel data comes from the Soli SDK, you can probably get more signal processing information from the Google Soli team. In our experiment setup, we use only one channel as the correlation between multi-channel doesn't help much with the accuracy - that is to say the same single channel setup is used across all the training and testing experiments.
datach is the number of the consecutive RDI frames of the same channel. It's a different concept comparison to the number of channels of a single frame.
JasperSrrzn
rohanghige
According to what I understand, each SoliData(###.h5) is transformed into 4 channels of images. However, in the network, it seems to be using only a single ch, which is the channel number assigned in main.lua. As I changed the assigned ch to 0, 2, 3 in main.lua the evaluation accuracy changes to 0.66988, 0.581660, 0.581513. Since 4 images(of channel 0~3) express a single data from SoliData, wouldn't all 4 channels be considered in training and evaluation process? If not, I am curious about what does each channel is indicating, and whether all 4 channels were used for training the network. Also, as implemented in ImageSeq:getImage(f) in imageseq.lua, is "datach" the number of consecutive RDI frames of the same channel that is stacked to form a part of input? It would be a great help if you help me with the confusion made by channels.