在对数据增强后的faster rcnn中进行训练时，出现这个错误，原因是在lib/roi_data_layer/layer.py中，会出现
inds = np.reshape(inds, (-1,2))
因此有效的roidb需为偶数。
参考https://github.com/rbgirshick/py-faster-rcnn/issues/61可处理数据奇偶问题
#inds = np.reshape(inds, (-1, 2))
#row_perm = np.random.permutation(np.arange(inds.shape[0]))
#inds = np.reshape(inds[row_perm, :], (-1,))
# consider the case when len(inds) is odd
isodd = len(inds) % 2
if not isodd:
inds = np.reshape(inds, (-1,2))
else:
tail = inds[-1]
inds = np.reshape(inds[:-1], (-1,2))
row_perm = np.random.permutation(np.arange(inds.shape[0]))
inds = np.reshape(inds[row_perm, :], (-1,))
if isodd:
inds = np.append(inds, tail)
self._perm = inds
ValueError: total size of new array must be unchanged
原文：http://www.cnblogs.com/573177885qq/p/7372630.html