def generate_inner_products_using_ip_split(net, step, filler):
bottom_name = "dropout%d" % step
generate_ip_split_layers(net, bottom_name, 250, "ip_conf%d" % step, 2)
generate_ip_split_layers(net, bottom_name, 250,
"ip_bbox_unscaled%d" % step, 4)
net.f(
Power("ip_bbox%d" % step,
scale=100,
bottoms=["ip_bbox_unscaled%d" % step]))
net.f(Softmax("ip_soft_conf%d" % step, bottoms=["ip_conf%d" % step]))
def generate_intermediate_layers(net):
"""Takes the output from the decapitated googlenet and transforms the output
from a NxCxWxH to (NxWxH)xCx1x1 that is used as input for the lstm layers.
N = batch size, C = channels, W = grid width, H = grid height."""
net.f(Convolution("post_fc7_conv", bottoms=["inception_final_output"],
param_lr_mults=[1., 2.], param_decay_mults=[0., 0.],
num_output=1024, kernel_dim=(1, 1),
weight_filler=Filler("gaussian", 0.005),
bias_filler=Filler("constant", 0.)))
net.f(Power("lstm_fc7_conv", scale=0.01, bottoms=["post_fc7_conv"]))
net.f(Transpose("lstm_input", bottoms=["lstm_fc7_conv"]))
def generate_inner_products(net, step, filler):
"""Inner products are fully connected layers. They generate
the final regressions for the confidence (ip_soft_conf),
and the bounding boxes (ip_bbox)"""
net.f(InnerProduct("ip_conf%d" % step, 2, bottoms=["dropout%d" % step],
output_4d=True,
weight_filler=filler))
net.f(InnerProduct("ip_bbox_unscaled%d" % step, 4,
bottoms=["dropout%d" % step], output_4d=True,
weight_filler=filler))
net.f(Power("ip_bbox%d" % step, scale=100,
bottoms=["ip_bbox_unscaled%d" % step]))
net.f(Softmax("ip_soft_conf%d" % step, bottoms=["ip_conf%d"%step]))
