print('output image shape:', out_img.shape)
d2l.plt.imshow(out_img.asnumpy())
#1x1卷积层采用Xavier随机初始化
net[-1].initialize(init.Constant(bilinear_kernel(num_classes, num_classes,
64)))
net[-2].initialize(init=init.Xavier())
#9.10.4-读取数据集
crop_size, batch_size, colormap2label = (320, 480), 1, nd.zeros(256**3) # 32
for i, cm in enumerate(d2l.VOC_COLORMAP):
colormap2label[(cm[0] * 256 + cm[1]) * 256 + cm[2]] = i
voc_dir = d2l.download_voc_pascal(data_dir='../data')
num_workers = 0 if sys.platform.startswith('win32') else 4
train_iter = gdata.DataLoader(d2l.VOCSegDataset(True, crop_size, voc_dir,
colormap2label),
batch_size,
shuffle=True,
last_batch='discard',
num_workers=num_workers)
test_iter = gdata.DataLoader(d2l.VOCSegDataset(False, crop_size, voc_dir,
colormap2label),
batch_size,
last_batch='discard',
num_workers=num_workers)
#9.10.5-训练模型
print('try_all_gpus')
ctx = d2l.try_all_gpus()
loss = gloss.SoftmaxCrossEntropyLoss(axis=1)
net.collect_params().reset_ctx(ctx)
for i, cm in enumerate(d2l.VOC_COLORMAP):
colormap2label[(cm[0] * 256 + cm[1]) * 256 + cm[2]] = i
#voc_dir = d2l.download_voc_pascal(data_dir='../data')
voc_dir = 'VOCdevkit/VOC2012'
num_workers = 0 if sys.platform.startswith('win32') else 4
train_iter = gdata.DataLoader(
d2l.VOCSegDataset(True, crop_size, voc_dir, colormap2label), batch_size,
shuffle=True, last_batch='discard', num_workers=num_workers)
test_iter = gdata.DataLoader(
d2l.VOCSegDataset(False, crop_size, voc_dir, colormap2label), batch_size,
last_batch='discard', num_workers=num_workers)
ctx = d2l.try_all_gpus()
loss = gloss.SoftmaxCrossEntropyLoss(axis=1)
net.collect_params().reset_ctx(ctx)