Open zimenglan-sysu-512 opened 3 years ago
hi @tancik since the
fc layer
indecoder
is too larger (1313 128 * 500 = 10,816,000), so i want use a conv(dim=1024) to increase the dimension and then usetf.reduce_mean(x, axis=[1, 2])
, but in the first 1w iterations, thebit_acc
always gets around 0.5, and thestr_acc
gets zero. so how to reduce the params for thefc layer
, at the same time, it can get good results?
Hey, is this problem solved? @zimenglan-sysu-512
hi @tancik since the
fc layer
indecoder
is too larger (1313 128 * 500 = 10,816,000), so i want use a conv(dim=1024) to increase the dimension and then usetf.reduce_mean(x, axis=[1, 2])
, but in the first 1w iterations, thebit_acc
always gets around 0.5, and thestr_acc
gets zero. so how to reduce the params for thefc layer
, at the same time, it can get good results?