PaddlePaddle_code

PaddlePaddle_code copied to clipboard

Performance issue in 3.image_classification/cifar_tf/cifar10_eval.py (by P3)

1

Hello! I've found a performance issue in cifar10_eval.py: `with tf.Session() as sess`[(here)](https://github.com/huxiaoman7/PaddlePaddle_code/blob/1ae81dcf9f7e32fb15eea8c4257ba02242cf6171/3.image_classification/cifar_tf/cifar10_eval.py#L76) is defined in the function `eval_once`[(here)](https://github.com/huxiaoman7/PaddlePaddle_code/blob/1ae81dcf9f7e32fb15eea8c4257ba02242cf6171/3.image_classification/cifar_tf/cifar10_eval.py#L67) which is repeatedly called in the loop `while True`[(here)](https://github.com/huxiaoman7/PaddlePaddle_code/blob/1ae81dcf9f7e32fb15eea8c4257ba02242cf6171/3.image_classification/cifar_tf/cifar10_eval.py#L147). `tf.Session` being defined...

DLPerf

小姐姐，你好 原始的Alexnet是用来处理Imagenet的数据，那个数据尺寸和cifar-10的尺寸不一样，请问你是如何处理的？

xyl576807077