GitHub - linxd5/cifar10: Some experiences to cifar10 dataset

Split model into cifar10_model.py file, and change it into slim version.
The original version of cifar10_eval.py will come to an NOTNAMEERROR. So I change the saver to restore tf.all_variables()

Change the model according to the idea of AlexNet (3x3 filter, more deeper).
Too small stddev (1e-4) of weight initialization leads to gradient vanish problem.
Remove batch_normaliztion leads to gradient vanish/explore problem (deep network).

Name		Name	Last commit message	Last commit date
Latest commit History 7 Commits
BUILD		BUILD
README.md		README.md
__init__.py		__init__.py
__init__.pyc		__init__.pyc
cifar10.py		cifar10.py
cifar10.pyc		cifar10.pyc
cifar10_eval.py		cifar10_eval.py
cifar10_input.py		cifar10_input.py
cifar10_input.pyc		cifar10_input.pyc
cifar10_input_test.py		cifar10_input_test.py
cifar10_model.py		cifar10_model.py
cifar10_model.pyc		cifar10_model.pyc
cifar10_model_pre.py		cifar10_model_pre.py
cifar10_model_pre.pyc		cifar10_model_pre.pyc
cifar10_train.py		cifar10_train.py
log.txt		log.txt

linxd5/cifar10