論文：Kwang Moo Yi, Eduard Trulls, Vincent Lepetit, Pascal Fua, ” LIFT: Learned Invariant Feature Transform”, in ECCV 2016, https://arxiv.org/abs/1603.09114
LIFT github 程式碼： https://github.com/cvlab-epfl/LIFT

所需軟體及其正確版本: CUDA (8.0), cuDNN (5.1), python (2.7), theano (0.90), Lasagne (0.2.dev1), flufl.lock (2.4.1), 剩下的numpy, scipy, parse, h5py 版本沒有具體要求。同時需要強調的是，有要求的版本必須嚴格匹配。

一. 相關軟體包安裝

1.CUDA (8.0), cuDNN (5.1)

測試程式碼：

from theano import function, config, shared, sandbox
import theano.tensor as T
import numpy
import time

vlen = 10 * 30 * 768  # 10 x #cores x # threads per core
iters = 1000

rng = numpy.random.RandomState(22)
x = shared(numpy.asarray(rng.rand(vlen), config.floatX)
 
)
f = function([], T.exp(x))
print(f.maker.fgraph.toposort())
t0 = time.time()
for i in range(iters):
    r = f()
t1 = time.time()
print("Looping %d times took %f seconds" % (iters, t1 - t0))
print("Result is %s" % (r,))
if numpy.any([isinstance(x.op, T.Elemwise) for x in f.maker.fgraph.toposort()]):
    print
 
('Used the cpu')
else:
    print('Used the gpu')

2.theano (0.90)

$ conda install theano=0.9

3.Lasagne (0.2.dev1)

$ conda install -c http://conda.anaconda.org/toli lasagne

4.flufl.lock (2.4.1)

$ pip install flufl.lock==2.4.1

二. 修改theano中少量程式碼

$ vim  /home/hansry/anaconda2/envs/py27/lib/python2.7/site-packages/lasagne/layers/pool.py

將 from theano.tensor.signal import downsampled 改為 from theano.tensor.signal.pool import pool_2d 並將下文中出現 downsample.max_pool 改為 pool_2d

至此，可正常執行LIFT程式碼。其實操作不難，主要是作者沒有將所需版本闡述清楚，所以才留下這麼多坑。只要一個版本不對，大量的報錯讓人無從下手。

有問題可私聊或者在下面留言！！