uint8 利用cast层转换为float32的疑问

[dyning]

@panyx0718

程序如下，初始化一个float32的numpy，然后通过numpy转为uint8，之后输入到paddle，通过cast转回float32，但是最终转换后的值和原始float32的值有diff，不知什么原因？
输出：
data numpy float32: [-148. -148. -148.]
data numpy uint8: [108 108 108]
paddle float32: [108. 108. 108.]

程序：
import paddle
import paddle.fluid as fluid
data = fluid.layers.data(name='data', shape=[3, 224, 224], dtype='uint8')
data_fl = fluid.layers.cast(x=data, dtype='float32')
#fluid.layers.Print(data_fl, summarize=10)
place = fluid.CUDAPlace(0)
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
img_blob = np.zeros((1, 224, 224, 3), dtype = np.float32)
img_blob = img_blob - 148
print "data numpy float32:", img_blob[0, 0, 0, 0:10]
img_blob = img_blob.astype(np.uint8)
print "data numpy uint8:", img_blob[0, 0, 0, 0:10]
img_blob_data = fluid.LoDTensor()
img_blob_data.set(img_blob, place)
result = exe.run(fluid.default_main_program(), feed={'data':img_blob_data}, fetch_list=[data_fl.name])
print "paddle float32:", result[0][0, 0, 0, 0:10]

[panyx0718]

uint8表达不了负数，溢出了

[dyning]

OK，了解，我试了numpy也转不回来。谢谢。
import paddle
import paddle.fluid as fluid
data = fluid.layers.data(name='data', shape=[3, 224, 224], dtype='uint8')
data_fl = fluid.layers.cast(x=data, dtype='float32')
#fluid.layers.Print(data_fl, summarize=10)
place = fluid.CUDAPlace(0)
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
img_blob = np.zeros((1, 224, 224, 3), dtype = np.float32)
img_blob = img_blob - 148
print "data numpy float32:", img_blob[0, 0, 0, 0:10]
img_blob = img_blob.astype(np.uint8)
print "data numpy uint8:", img_blob[0, 0, 0, 0:10]
img_blob = img_blob.astype(np.float32)
print "data convert numpy float32:", img_blob[0, 0, 0, 0:10]
img_blob_data = fluid.LoDTensor()
img_blob_data.set(img_blob, place)
result = exe.run(fluid.default_main_program(), feed={'data':img_blob_data}, fetch_list=[data_fl.name])
print "paddle float32:", result[0][0, 0, 0, 0:10]