backbone/stem/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[7, 7, 3, 64], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stem/norm1/scale: constant(1.0)
backbone/stem/norm1/bias: constant(0.0)
backbone/stem/norm1/moving_mean: constant(0.0)
backbone/stem/norm1/moving_variance: constant(1.0)
backbone/stage0/block0/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 64, 64], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage0/block0/norm1/scale: constant(1.0)
backbone/stage0/block0/norm1/bias: constant(0.0)
backbone/stage0/block0/norm1/moving_mean: constant(0.0)
backbone/stage0/block0/norm1/moving_variance: constant(1.0)
backbone/stage0/block0/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 64, 64], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage0/block0/norm2/scale: constant(1.0)
backbone/stage0/block0/norm2/bias: constant(0.0)
backbone/stage0/block0/norm2/moving_mean: constant(0.0)
backbone/stage0/block0/norm2/moving_variance: constant(1.0)
backbone/stage0/block0/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 64, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage0/block0/norm3/scale: constant(1.0)
backbone/stage0/block0/norm3/bias: constant(0.0)
backbone/stage0/block0/norm3/moving_mean: constant(0.0)
backbone/stage0/block0/norm3/moving_variance: constant(1.0)
backbone/stage0/block0/downsample/conv/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 64, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage0/block0/downsample/norm/scale: constant(1.0)
backbone/stage0/block0/downsample/norm/bias: constant(0.0)
backbone/stage0/block0/downsample/norm/moving_mean: constant(0.0)
backbone/stage0/block0/downsample/norm/moving_variance: constant(1.0)
backbone/stage0/block1/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 64], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage0/block1/norm1/scale: constant(1.0)
backbone/stage0/block1/norm1/bias: constant(0.0)
backbone/stage0/block1/norm1/moving_mean: constant(0.0)
backbone/stage0/block1/norm1/moving_variance: constant(1.0)
backbone/stage0/block1/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 64, 64], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage0/block1/norm2/scale: constant(1.0)
backbone/stage0/block1/norm2/bias: constant(0.0)
backbone/stage0/block1/norm2/moving_mean: constant(0.0)
backbone/stage0/block1/norm2/moving_variance: constant(1.0)
backbone/stage0/block1/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 64, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage0/block1/norm3/scale: constant(1.0)
backbone/stage0/block1/norm3/bias: constant(0.0)
backbone/stage0/block1/norm3/moving_mean: constant(0.0)
backbone/stage0/block1/norm3/moving_variance: constant(1.0)
backbone/stage0/block2/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 64], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage0/block2/norm1/scale: constant(1.0)
backbone/stage0/block2/norm1/bias: constant(0.0)
backbone/stage0/block2/norm1/moving_mean: constant(0.0)
backbone/stage0/block2/norm1/moving_variance: constant(1.0)
backbone/stage0/block2/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 64, 64], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage0/block2/norm2/scale: constant(1.0)
backbone/stage0/block2/norm2/bias: constant(0.0)
backbone/stage0/block2/norm2/moving_mean: constant(0.0)
backbone/stage0/block2/norm2/moving_variance: constant(1.0)
backbone/stage0/block2/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 64, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage0/block2/norm3/scale: constant(1.0)
backbone/stage0/block2/norm3/bias: constant(0.0)
backbone/stage0/block2/norm3/moving_mean: constant(0.0)
backbone/stage0/block2/norm3/moving_variance: constant(1.0)
backbone/stage1/block0/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 128], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block0/norm1/scale: constant(1.0)
backbone/stage1/block0/norm1/bias: constant(0.0)
backbone/stage1/block0/norm1/moving_mean: constant(0.0)
backbone/stage1/block0/norm1/moving_variance: constant(1.0)
backbone/stage1/block0/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 128, 128], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block0/norm2/scale: constant(1.0)
backbone/stage1/block0/norm2/bias: constant(0.0)
backbone/stage1/block0/norm2/moving_mean: constant(0.0)
backbone/stage1/block0/norm2/moving_variance: constant(1.0)
backbone/stage1/block0/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 128, 512], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block0/norm3/scale: constant(1.0)
backbone/stage1/block0/norm3/bias: constant(0.0)
backbone/stage1/block0/norm3/moving_mean: constant(0.0)
backbone/stage1/block0/norm3/moving_variance: constant(1.0)
backbone/stage1/block0/downsample/conv/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 512], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block0/downsample/norm/scale: constant(1.0)
backbone/stage1/block0/downsample/norm/bias: constant(0.0)
backbone/stage1/block0/downsample/norm/moving_mean: constant(0.0)
backbone/stage1/block0/downsample/norm/moving_variance: constant(1.0)
backbone/stage1/block1/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 512, 128], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block1/norm1/scale: constant(1.0)
backbone/stage1/block1/norm1/bias: constant(0.0)
backbone/stage1/block1/norm1/moving_mean: constant(0.0)
backbone/stage1/block1/norm1/moving_variance: constant(1.0)
backbone/stage1/block1/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 128, 128], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block1/norm2/scale: constant(1.0)
backbone/stage1/block1/norm2/bias: constant(0.0)
backbone/stage1/block1/norm2/moving_mean: constant(0.0)
backbone/stage1/block1/norm2/moving_variance: constant(1.0)
backbone/stage1/block1/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 128, 512], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block1/norm3/scale: constant(1.0)
backbone/stage1/block1/norm3/bias: constant(0.0)
backbone/stage1/block1/norm3/moving_mean: constant(0.0)
backbone/stage1/block1/norm3/moving_variance: constant(1.0)
backbone/stage1/block2/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 512, 128], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block2/norm1/scale: constant(1.0)
backbone/stage1/block2/norm1/bias: constant(0.0)
backbone/stage1/block2/norm1/moving_mean: constant(0.0)
backbone/stage1/block2/norm1/moving_variance: constant(1.0)
backbone/stage1/block2/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 128, 128], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block2/norm2/scale: constant(1.0)
backbone/stage1/block2/norm2/bias: constant(0.0)
backbone/stage1/block2/norm2/moving_mean: constant(0.0)
backbone/stage1/block2/norm2/moving_variance: constant(1.0)
backbone/stage1/block2/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 128, 512], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block2/norm3/scale: constant(1.0)
backbone/stage1/block2/norm3/bias: constant(0.0)
backbone/stage1/block2/norm3/moving_mean: constant(0.0)
backbone/stage1/block2/norm3/moving_variance: constant(1.0)
backbone/stage1/block3/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 512, 128], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block3/norm1/scale: constant(1.0)
backbone/stage1/block3/norm1/bias: constant(0.0)
backbone/stage1/block3/norm1/moving_mean: constant(0.0)
backbone/stage1/block3/norm1/moving_variance: constant(1.0)
backbone/stage1/block3/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 128, 128], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block3/norm2/scale: constant(1.0)
backbone/stage1/block3/norm2/bias: constant(0.0)
backbone/stage1/block3/norm2/moving_mean: constant(0.0)
backbone/stage1/block3/norm2/moving_variance: constant(1.0)
backbone/stage1/block3/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 128, 512], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage1/block3/norm3/scale: constant(1.0)
backbone/stage1/block3/norm3/bias: constant(0.0)
backbone/stage1/block3/norm3/moving_mean: constant(0.0)
backbone/stage1/block3/norm3/moving_variance: constant(1.0)
backbone/stage2/block0/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 512, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block0/norm1/scale: constant(1.0)
backbone/stage2/block0/norm1/bias: constant(0.0)
backbone/stage2/block0/norm1/moving_mean: constant(0.0)
backbone/stage2/block0/norm1/moving_variance: constant(1.0)
backbone/stage2/block0/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block0/norm2/scale: constant(1.0)
backbone/stage2/block0/norm2/bias: constant(0.0)
backbone/stage2/block0/norm2/moving_mean: constant(0.0)
backbone/stage2/block0/norm2/moving_variance: constant(1.0)
backbone/stage2/block0/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block0/norm3/scale: constant(1.0)
backbone/stage2/block0/norm3/bias: constant(0.0)
backbone/stage2/block0/norm3/moving_mean: constant(0.0)
backbone/stage2/block0/norm3/moving_variance: constant(1.0)
backbone/stage2/block0/downsample/conv/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 512, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block0/downsample/norm/scale: constant(1.0)
backbone/stage2/block0/downsample/norm/bias: constant(0.0)
backbone/stage2/block0/downsample/norm/moving_mean: constant(0.0)
backbone/stage2/block0/downsample/norm/moving_variance: constant(1.0)
backbone/stage2/block1/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block1/norm1/scale: constant(1.0)
backbone/stage2/block1/norm1/bias: constant(0.0)
backbone/stage2/block1/norm1/moving_mean: constant(0.0)
backbone/stage2/block1/norm1/moving_variance: constant(1.0)
backbone/stage2/block1/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block1/norm2/scale: constant(1.0)
backbone/stage2/block1/norm2/bias: constant(0.0)
backbone/stage2/block1/norm2/moving_mean: constant(0.0)
backbone/stage2/block1/norm2/moving_variance: constant(1.0)
backbone/stage2/block1/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block1/norm3/scale: constant(1.0)
backbone/stage2/block1/norm3/bias: constant(0.0)
backbone/stage2/block1/norm3/moving_mean: constant(0.0)
backbone/stage2/block1/norm3/moving_variance: constant(1.0)
backbone/stage2/block2/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block2/norm1/scale: constant(1.0)
backbone/stage2/block2/norm1/bias: constant(0.0)
backbone/stage2/block2/norm1/moving_mean: constant(0.0)
backbone/stage2/block2/norm1/moving_variance: constant(1.0)
backbone/stage2/block2/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block2/norm2/scale: constant(1.0)
backbone/stage2/block2/norm2/bias: constant(0.0)
backbone/stage2/block2/norm2/moving_mean: constant(0.0)
backbone/stage2/block2/norm2/moving_variance: constant(1.0)
backbone/stage2/block2/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block2/norm3/scale: constant(1.0)
backbone/stage2/block2/norm3/bias: constant(0.0)
backbone/stage2/block2/norm3/moving_mean: constant(0.0)
backbone/stage2/block2/norm3/moving_variance: constant(1.0)
backbone/stage2/block3/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block3/norm1/scale: constant(1.0)
backbone/stage2/block3/norm1/bias: constant(0.0)
backbone/stage2/block3/norm1/moving_mean: constant(0.0)
backbone/stage2/block3/norm1/moving_variance: constant(1.0)
backbone/stage2/block3/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block3/norm2/scale: constant(1.0)
backbone/stage2/block3/norm2/bias: constant(0.0)
backbone/stage2/block3/norm2/moving_mean: constant(0.0)
backbone/stage2/block3/norm2/moving_variance: constant(1.0)
backbone/stage2/block3/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block3/norm3/scale: constant(1.0)
backbone/stage2/block3/norm3/bias: constant(0.0)
backbone/stage2/block3/norm3/moving_mean: constant(0.0)
backbone/stage2/block3/norm3/moving_variance: constant(1.0)
backbone/stage2/block4/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block4/norm1/scale: constant(1.0)
backbone/stage2/block4/norm1/bias: constant(0.0)
backbone/stage2/block4/norm1/moving_mean: constant(0.0)
backbone/stage2/block4/norm1/moving_variance: constant(1.0)
backbone/stage2/block4/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block4/norm2/scale: constant(1.0)
backbone/stage2/block4/norm2/bias: constant(0.0)
backbone/stage2/block4/norm2/moving_mean: constant(0.0)
backbone/stage2/block4/norm2/moving_variance: constant(1.0)
backbone/stage2/block4/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block4/norm3/scale: constant(1.0)
backbone/stage2/block4/norm3/bias: constant(0.0)
backbone/stage2/block4/norm3/moving_mean: constant(0.0)
backbone/stage2/block4/norm3/moving_variance: constant(1.0)
backbone/stage2/block5/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block5/norm1/scale: constant(1.0)
backbone/stage2/block5/norm1/bias: constant(0.0)
backbone/stage2/block5/norm1/moving_mean: constant(0.0)
backbone/stage2/block5/norm1/moving_variance: constant(1.0)
backbone/stage2/block5/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block5/norm2/scale: constant(1.0)
backbone/stage2/block5/norm2/bias: constant(0.0)
backbone/stage2/block5/norm2/moving_mean: constant(0.0)
backbone/stage2/block5/norm2/moving_variance: constant(1.0)
backbone/stage2/block5/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block5/norm3/scale: constant(1.0)
backbone/stage2/block5/norm3/bias: constant(0.0)
backbone/stage2/block5/norm3/moving_mean: constant(0.0)
backbone/stage2/block5/norm3/moving_variance: constant(1.0)
backbone/stage2/block6/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block6/norm1/scale: constant(1.0)
backbone/stage2/block6/norm1/bias: constant(0.0)
backbone/stage2/block6/norm1/moving_mean: constant(0.0)
backbone/stage2/block6/norm1/moving_variance: constant(1.0)
backbone/stage2/block6/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block6/norm2/scale: constant(1.0)
backbone/stage2/block6/norm2/bias: constant(0.0)
backbone/stage2/block6/norm2/moving_mean: constant(0.0)
backbone/stage2/block6/norm2/moving_variance: constant(1.0)
backbone/stage2/block6/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block6/norm3/scale: constant(1.0)
backbone/stage2/block6/norm3/bias: constant(0.0)
backbone/stage2/block6/norm3/moving_mean: constant(0.0)
backbone/stage2/block6/norm3/moving_variance: constant(1.0)
backbone/stage2/block7/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block7/norm1/scale: constant(1.0)
backbone/stage2/block7/norm1/bias: constant(0.0)
backbone/stage2/block7/norm1/moving_mean: constant(0.0)
backbone/stage2/block7/norm1/moving_variance: constant(1.0)
backbone/stage2/block7/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block7/norm2/scale: constant(1.0)
backbone/stage2/block7/norm2/bias: constant(0.0)
backbone/stage2/block7/norm2/moving_mean: constant(0.0)
backbone/stage2/block7/norm2/moving_variance: constant(1.0)
backbone/stage2/block7/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block7/norm3/scale: constant(1.0)
backbone/stage2/block7/norm3/bias: constant(0.0)
backbone/stage2/block7/norm3/moving_mean: constant(0.0)
backbone/stage2/block7/norm3/moving_variance: constant(1.0)
backbone/stage2/block8/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block8/norm1/scale: constant(1.0)
backbone/stage2/block8/norm1/bias: constant(0.0)
backbone/stage2/block8/norm1/moving_mean: constant(0.0)
backbone/stage2/block8/norm1/moving_variance: constant(1.0)
backbone/stage2/block8/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block8/norm2/scale: constant(1.0)
backbone/stage2/block8/norm2/bias: constant(0.0)
backbone/stage2/block8/norm2/moving_mean: constant(0.0)
backbone/stage2/block8/norm2/moving_variance: constant(1.0)
backbone/stage2/block8/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block8/norm3/scale: constant(1.0)
backbone/stage2/block8/norm3/bias: constant(0.0)
backbone/stage2/block8/norm3/moving_mean: constant(0.0)
backbone/stage2/block8/norm3/moving_variance: constant(1.0)
backbone/stage2/block9/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block9/norm1/scale: constant(1.0)
backbone/stage2/block9/norm1/bias: constant(0.0)
backbone/stage2/block9/norm1/moving_mean: constant(0.0)
backbone/stage2/block9/norm1/moving_variance: constant(1.0)
backbone/stage2/block9/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block9/norm2/scale: constant(1.0)
backbone/stage2/block9/norm2/bias: constant(0.0)
backbone/stage2/block9/norm2/moving_mean: constant(0.0)
backbone/stage2/block9/norm2/moving_variance: constant(1.0)
backbone/stage2/block9/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block9/norm3/scale: constant(1.0)
backbone/stage2/block9/norm3/bias: constant(0.0)
backbone/stage2/block9/norm3/moving_mean: constant(0.0)
backbone/stage2/block9/norm3/moving_variance: constant(1.0)
backbone/stage2/block10/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block10/norm1/scale: constant(1.0)
backbone/stage2/block10/norm1/bias: constant(0.0)
backbone/stage2/block10/norm1/moving_mean: constant(0.0)
backbone/stage2/block10/norm1/moving_variance: constant(1.0)
backbone/stage2/block10/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block10/norm2/scale: constant(1.0)
backbone/stage2/block10/norm2/bias: constant(0.0)
backbone/stage2/block10/norm2/moving_mean: constant(0.0)
backbone/stage2/block10/norm2/moving_variance: constant(1.0)
backbone/stage2/block10/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block10/norm3/scale: constant(1.0)
backbone/stage2/block10/norm3/bias: constant(0.0)
backbone/stage2/block10/norm3/moving_mean: constant(0.0)
backbone/stage2/block10/norm3/moving_variance: constant(1.0)
backbone/stage2/block11/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block11/norm1/scale: constant(1.0)
backbone/stage2/block11/norm1/bias: constant(0.0)
backbone/stage2/block11/norm1/moving_mean: constant(0.0)
backbone/stage2/block11/norm1/moving_variance: constant(1.0)
backbone/stage2/block11/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block11/norm2/scale: constant(1.0)
backbone/stage2/block11/norm2/bias: constant(0.0)
backbone/stage2/block11/norm2/moving_mean: constant(0.0)
backbone/stage2/block11/norm2/moving_variance: constant(1.0)
backbone/stage2/block11/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block11/norm3/scale: constant(1.0)
backbone/stage2/block11/norm3/bias: constant(0.0)
backbone/stage2/block11/norm3/moving_mean: constant(0.0)
backbone/stage2/block11/norm3/moving_variance: constant(1.0)
backbone/stage2/block12/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block12/norm1/scale: constant(1.0)
backbone/stage2/block12/norm1/bias: constant(0.0)
backbone/stage2/block12/norm1/moving_mean: constant(0.0)
backbone/stage2/block12/norm1/moving_variance: constant(1.0)
backbone/stage2/block12/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block12/norm2/scale: constant(1.0)
backbone/stage2/block12/norm2/bias: constant(0.0)
backbone/stage2/block12/norm2/moving_mean: constant(0.0)
backbone/stage2/block12/norm2/moving_variance: constant(1.0)
backbone/stage2/block12/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block12/norm3/scale: constant(1.0)
backbone/stage2/block12/norm3/bias: constant(0.0)
backbone/stage2/block12/norm3/moving_mean: constant(0.0)
backbone/stage2/block12/norm3/moving_variance: constant(1.0)
backbone/stage2/block13/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block13/norm1/scale: constant(1.0)
backbone/stage2/block13/norm1/bias: constant(0.0)
backbone/stage2/block13/norm1/moving_mean: constant(0.0)
backbone/stage2/block13/norm1/moving_variance: constant(1.0)
backbone/stage2/block13/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block13/norm2/scale: constant(1.0)
backbone/stage2/block13/norm2/bias: constant(0.0)
backbone/stage2/block13/norm2/moving_mean: constant(0.0)
backbone/stage2/block13/norm2/moving_variance: constant(1.0)
backbone/stage2/block13/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block13/norm3/scale: constant(1.0)
backbone/stage2/block13/norm3/bias: constant(0.0)
backbone/stage2/block13/norm3/moving_mean: constant(0.0)
backbone/stage2/block13/norm3/moving_variance: constant(1.0)
backbone/stage2/block14/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block14/norm1/scale: constant(1.0)
backbone/stage2/block14/norm1/bias: constant(0.0)
backbone/stage2/block14/norm1/moving_mean: constant(0.0)
backbone/stage2/block14/norm1/moving_variance: constant(1.0)
backbone/stage2/block14/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block14/norm2/scale: constant(1.0)
backbone/stage2/block14/norm2/bias: constant(0.0)
backbone/stage2/block14/norm2/moving_mean: constant(0.0)
backbone/stage2/block14/norm2/moving_variance: constant(1.0)
backbone/stage2/block14/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block14/norm3/scale: constant(1.0)
backbone/stage2/block14/norm3/bias: constant(0.0)
backbone/stage2/block14/norm3/moving_mean: constant(0.0)
backbone/stage2/block14/norm3/moving_variance: constant(1.0)
backbone/stage2/block15/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block15/norm1/scale: constant(1.0)
backbone/stage2/block15/norm1/bias: constant(0.0)
backbone/stage2/block15/norm1/moving_mean: constant(0.0)
backbone/stage2/block15/norm1/moving_variance: constant(1.0)
backbone/stage2/block15/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block15/norm2/scale: constant(1.0)
backbone/stage2/block15/norm2/bias: constant(0.0)
backbone/stage2/block15/norm2/moving_mean: constant(0.0)
backbone/stage2/block15/norm2/moving_variance: constant(1.0)
backbone/stage2/block15/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block15/norm3/scale: constant(1.0)
backbone/stage2/block15/norm3/bias: constant(0.0)
backbone/stage2/block15/norm3/moving_mean: constant(0.0)
backbone/stage2/block15/norm3/moving_variance: constant(1.0)
backbone/stage2/block16/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block16/norm1/scale: constant(1.0)
backbone/stage2/block16/norm1/bias: constant(0.0)
backbone/stage2/block16/norm1/moving_mean: constant(0.0)
backbone/stage2/block16/norm1/moving_variance: constant(1.0)
backbone/stage2/block16/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block16/norm2/scale: constant(1.0)
backbone/stage2/block16/norm2/bias: constant(0.0)
backbone/stage2/block16/norm2/moving_mean: constant(0.0)
backbone/stage2/block16/norm2/moving_variance: constant(1.0)
backbone/stage2/block16/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block16/norm3/scale: constant(1.0)
backbone/stage2/block16/norm3/bias: constant(0.0)
backbone/stage2/block16/norm3/moving_mean: constant(0.0)
backbone/stage2/block16/norm3/moving_variance: constant(1.0)
backbone/stage2/block17/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block17/norm1/scale: constant(1.0)
backbone/stage2/block17/norm1/bias: constant(0.0)
backbone/stage2/block17/norm1/moving_mean: constant(0.0)
backbone/stage2/block17/norm1/moving_variance: constant(1.0)
backbone/stage2/block17/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block17/norm2/scale: constant(1.0)
backbone/stage2/block17/norm2/bias: constant(0.0)
backbone/stage2/block17/norm2/moving_mean: constant(0.0)
backbone/stage2/block17/norm2/moving_variance: constant(1.0)
backbone/stage2/block17/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block17/norm3/scale: constant(1.0)
backbone/stage2/block17/norm3/bias: constant(0.0)
backbone/stage2/block17/norm3/moving_mean: constant(0.0)
backbone/stage2/block17/norm3/moving_variance: constant(1.0)
backbone/stage2/block18/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block18/norm1/scale: constant(1.0)
backbone/stage2/block18/norm1/bias: constant(0.0)
backbone/stage2/block18/norm1/moving_mean: constant(0.0)
backbone/stage2/block18/norm1/moving_variance: constant(1.0)
backbone/stage2/block18/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block18/norm2/scale: constant(1.0)
backbone/stage2/block18/norm2/bias: constant(0.0)
backbone/stage2/block18/norm2/moving_mean: constant(0.0)
backbone/stage2/block18/norm2/moving_variance: constant(1.0)
backbone/stage2/block18/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block18/norm3/scale: constant(1.0)
backbone/stage2/block18/norm3/bias: constant(0.0)
backbone/stage2/block18/norm3/moving_mean: constant(0.0)
backbone/stage2/block18/norm3/moving_variance: constant(1.0)
backbone/stage2/block19/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block19/norm1/scale: constant(1.0)
backbone/stage2/block19/norm1/bias: constant(0.0)
backbone/stage2/block19/norm1/moving_mean: constant(0.0)
backbone/stage2/block19/norm1/moving_variance: constant(1.0)
backbone/stage2/block19/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block19/norm2/scale: constant(1.0)
backbone/stage2/block19/norm2/bias: constant(0.0)
backbone/stage2/block19/norm2/moving_mean: constant(0.0)
backbone/stage2/block19/norm2/moving_variance: constant(1.0)
backbone/stage2/block19/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block19/norm3/scale: constant(1.0)
backbone/stage2/block19/norm3/bias: constant(0.0)
backbone/stage2/block19/norm3/moving_mean: constant(0.0)
backbone/stage2/block19/norm3/moving_variance: constant(1.0)
backbone/stage2/block20/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block20/norm1/scale: constant(1.0)
backbone/stage2/block20/norm1/bias: constant(0.0)
backbone/stage2/block20/norm1/moving_mean: constant(0.0)
backbone/stage2/block20/norm1/moving_variance: constant(1.0)
backbone/stage2/block20/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block20/norm2/scale: constant(1.0)
backbone/stage2/block20/norm2/bias: constant(0.0)
backbone/stage2/block20/norm2/moving_mean: constant(0.0)
backbone/stage2/block20/norm2/moving_variance: constant(1.0)
backbone/stage2/block20/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block20/norm3/scale: constant(1.0)
backbone/stage2/block20/norm3/bias: constant(0.0)
backbone/stage2/block20/norm3/moving_mean: constant(0.0)
backbone/stage2/block20/norm3/moving_variance: constant(1.0)
backbone/stage2/block21/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block21/norm1/scale: constant(1.0)
backbone/stage2/block21/norm1/bias: constant(0.0)
backbone/stage2/block21/norm1/moving_mean: constant(0.0)
backbone/stage2/block21/norm1/moving_variance: constant(1.0)
backbone/stage2/block21/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block21/norm2/scale: constant(1.0)
backbone/stage2/block21/norm2/bias: constant(0.0)
backbone/stage2/block21/norm2/moving_mean: constant(0.0)
backbone/stage2/block21/norm2/moving_variance: constant(1.0)
backbone/stage2/block21/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block21/norm3/scale: constant(1.0)
backbone/stage2/block21/norm3/bias: constant(0.0)
backbone/stage2/block21/norm3/moving_mean: constant(0.0)
backbone/stage2/block21/norm3/moving_variance: constant(1.0)
backbone/stage2/block22/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block22/norm1/scale: constant(1.0)
backbone/stage2/block22/norm1/bias: constant(0.0)
backbone/stage2/block22/norm1/moving_mean: constant(0.0)
backbone/stage2/block22/norm1/moving_variance: constant(1.0)
backbone/stage2/block22/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 256, 256], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block22/norm2/scale: constant(1.0)
backbone/stage2/block22/norm2/bias: constant(0.0)
backbone/stage2/block22/norm2/moving_mean: constant(0.0)
backbone/stage2/block22/norm2/moving_variance: constant(1.0)
backbone/stage2/block22/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 256, 1024], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage2/block22/norm3/scale: constant(1.0)
backbone/stage2/block22/norm3/bias: constant(0.0)
backbone/stage2/block22/norm3/moving_mean: constant(0.0)
backbone/stage2/block22/norm3/moving_variance: constant(1.0)
backbone/stage3/block0/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 512], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage3/block0/norm1/scale: constant(1.0)
backbone/stage3/block0/norm1/bias: constant(0.0)
backbone/stage3/block0/norm1/moving_mean: constant(0.0)
backbone/stage3/block0/norm1/moving_variance: constant(1.0)
backbone/stage3/block0/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 512, 512], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage3/block0/norm2/scale: constant(1.0)
backbone/stage3/block0/norm2/bias: constant(0.0)
backbone/stage3/block0/norm2/moving_mean: constant(0.0)
backbone/stage3/block0/norm2/moving_variance: constant(1.0)
backbone/stage3/block0/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 512, 2048], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage3/block0/norm3/scale: constant(1.0)
backbone/stage3/block0/norm3/bias: constant(0.0)
backbone/stage3/block0/norm3/moving_mean: constant(0.0)
backbone/stage3/block0/norm3/moving_variance: constant(1.0)
backbone/stage3/block0/downsample/conv/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 1024, 2048], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage3/block0/downsample/norm/scale: constant(1.0)
backbone/stage3/block0/downsample/norm/bias: constant(0.0)
backbone/stage3/block0/downsample/norm/moving_mean: constant(0.0)
backbone/stage3/block0/downsample/norm/moving_variance: constant(1.0)
backbone/stage3/block1/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 2048, 512], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage3/block1/norm1/scale: constant(1.0)
backbone/stage3/block1/norm1/bias: constant(0.0)
backbone/stage3/block1/norm1/moving_mean: constant(0.0)
backbone/stage3/block1/norm1/moving_variance: constant(1.0)
backbone/stage3/block1/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 512, 512], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage3/block1/norm2/scale: constant(1.0)
backbone/stage3/block1/norm2/bias: constant(0.0)
backbone/stage3/block1/norm2/moving_mean: constant(0.0)
backbone/stage3/block1/norm2/moving_variance: constant(1.0)
backbone/stage3/block1/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 512, 2048], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage3/block1/norm3/scale: constant(1.0)
backbone/stage3/block1/norm3/bias: constant(0.0)
backbone/stage3/block1/norm3/moving_mean: constant(0.0)
backbone/stage3/block1/norm3/moving_variance: constant(1.0)
backbone/stage3/block2/conv1/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 2048, 512], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage3/block2/norm1/scale: constant(1.0)
backbone/stage3/block2/norm1/bias: constant(0.0)
backbone/stage3/block2/norm1/moving_mean: constant(0.0)
backbone/stage3/block2/norm1/moving_variance: constant(1.0)
backbone/stage3/block2/conv2/weight: normal(0, 1.4142135623730951 / fan_out), shape=[3, 3, 512, 512], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage3/block2/norm2/scale: constant(1.0)
backbone/stage3/block2/norm2/bias: constant(0.0)
backbone/stage3/block2/norm2/moving_mean: constant(0.0)
backbone/stage3/block2/norm2/moving_variance: constant(1.0)
backbone/stage3/block2/conv3/weight: normal(0, 1.4142135623730951 / fan_out), shape=[1, 1, 512, 2048], axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
backbone/stage3/block2/norm3/scale: constant(1.0)
backbone/stage3/block2/norm3/bias: constant(0.0)
backbone/stage3/block2/norm3/moving_mean: constant(0.0)
backbone/stage3/block2/norm3/moving_variance: constant(1.0)
classifier/weight: normal(0, 1.4142135623730951 / fan_out), shape=(2048, 1000), axes=FanAxes(in_axis=-2, out_axis=-1, batch_axis=())
classifier/bias: constant(0.0)