def transform_predict(data, data_shape=(3, 331, 331), resize=331):
ctx = data.context
im = resize_longer(data.asnumpy(), resize, data_shape)
im = nd.array(im, ctx=ctx)
im = im.astype('float32') / 255
auglist = [
# image.RandomGrayAug(1.0),
image.ColorNormalizeAug(mean=(0.417, 0.402, 0.366),
std=(0.081, 0.079, 0.080)),
]
for aug in auglist:
im = aug(im)
im = nd.transpose(im, (2, 0, 1))
return nd.stack(im)
def transform_val(data, label, data_shape=(3, 331, 331), resize=331):
# im = data.astype('float32') / 255,
ctx = data.context
im = resize_longer(data.asnumpy(), resize, data_shape)
im = nd.array(im, ctx=ctx)
im = im.astype('float32') / 255
auglist = [
# image.RandomGrayAug(1.0),
# image.ColorNormalizeAug(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)),
image.ColorNormalizeAug(mean=(0.417, 0.402, 0.366),
std=(0.081, 0.079, 0.080)),
]
for aug in auglist:
im = aug(im)
# im = im.clip(0, 1)
im = nd.transpose(im, (2, 0, 1))
return im, nd.array([label], ctx=ctx).asscalar()
def transform_train_DA3(data, label):
im = data.astype(np.float32) / 255
auglist = [image.RandomSizedCropAug(size=(32, 32), min_area=0.49, ratio=(0.5, 2))]
_aug = image.CreateAugmenter(data_shape=(3, 32, 32), resize=0,
rand_crop=False, rand_resize=False, rand_mirror=True,
# mean=np.array([0.4914, 0.4822, 0.4465]),
# std=np.array([0.2023, 0.1994, 0.2010]),
brightness=0.3, contrast=0.3, saturation=0.3, hue=0.3,
pca_noise=0.01, rand_gray=0, inter_method=2)
auglist.append(image.RandomOrderAug(_aug))
for aug in auglist:
im = aug(im)
if random.random() > random_clip_rate:
im = im.clip(0, 1)
_aug = image.ColorNormalizeAug(mean=np.array([0.4914, 0.4822, 0.4465]),
std=np.array([0.2023, 0.1994, 0.2010]), )
im = _aug(im)
im = nd.transpose(im, (2, 0, 1))
return (im, nd.array([label]).asscalar().astype('float32'))
def transform_train(data, label, data_shape=(3, 363, 363), resize=363):
# data [height, width, channel]
# im = data.astype('float32') / 255,
ctx = data.context
im = resize_longer(data.asnumpy(), resize, data_shape)
im = rotate_image(im)
im = nd.array(im, ctx=ctx)
im = im.astype('float32') / 255
auglist = [
image.RandomCropAug((331, 331)),
image.ColorJitterAug(0.3, 0.3, 0.3),
image.RandomGrayAug(0.5),
# image.ColorNormalizeAug(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)),
image.ColorNormalizeAug(mean=(0.417, 0.402, 0.366),
std=(0.081, 0.079, 0.080)),
]
for aug in auglist:
im = aug(im)
# im = im.clip(0, 1)
im = nd.transpose(im, (2, 0, 1))
return im, nd.array([label], ctx=ctx).asscalar()
import gc
from tqdm import tqdm
from sklearn.model_selection import train_test_split
# In[ ]:
# try run via gpu
ctx = mx.gpu()
# In[ ]:
transformers = [
# 强制resize成pretrain模型的输入大小
image.ForceResizeAug((224, 224)),
# 标准化处理
image.ColorNormalizeAug(mean=nd.array([0.485, 0.456, 0.406]),
std=nd.array([0.229, 0.224, 0.225]))
]
# Transoform image and label to our target format data
#
# 为避免过拟合，我们在这里使用`image.CreateAugmenter`来加强数据集。例如我们设`rand_mirror=True`即可随机对每张图片做镜面反转。我们也通过`mean`和`std`对彩色图像RGB三个通道分别做标准化。以下我们列举了该函数里的所有参数，这些参数都是可以调的。
# In[ ]:
train_augs_params = {
"resize": 1,
"rand_crop": False,
"rand_resize": False,
"rand_mirror": True,
"mean": np.array([0.4914, 0.4822, 0.4465]),
"std": np.array([0.2023, 0.1994, 0.2010]),
