Model

Info

Our model consists of

model = nn.Sequential()
model:add(nn.SpatialConvolution(3,6,5,5))
model:add(nn.SpatialMaxPooling(2,2,2,2))
model:add(nn.ReLU())
model:add(nn.SpatialConvolution(6,16,5,5))
model:add(nn.SpatialMaxPooling(2,2,2,2))
model:add(nn.ReLU())
model:add(nn.View(16*2*2))
model:add(nn.Linear(16*2*2,120))
model:add(nn.ReLU())
model:add(nn.Linear(120,80))
model:add(nn.ReLU())
model:add(nn.Linear(80,36))
model:add(nn.LogSoftMax())

Training

We have train our model upto 100 iterations.

Spatial Convolution

module = nn.SpatialConvolution(nInputPlane, nOutputPlane, kW, kH)

• kW: The kernel width of the convolution.
• kh: The kernel height of the convolution. Filter used is of size 5x5.

Spatial MaxPooling

module = nn.SpatialMaxPooling(kW, kH, dW, dH)

• kW: The kernel width of the convolution.
• kh: The kernel height of the convolution.
• dW: The step of the convolution in the width dimension.
• dh: The step of the convolution in the height dimension.

Relu

ReLU is defined asf(x) = max(0,x)

LogSoftMax

LogSoftmax is defined as f_i(x) = log(1/a exp(x_i)), where a = sum_j exp(x_j).