Even if the limit the gpu memory usage option is enabled, segmentation ball app run crash
# Limit the GPU memory usage.
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = FLAGS.gpu_memory_usage
config.gpu_options.allow_growth = True
even with your settings
flags.DEFINE_float('gpu_memory_usage', 0.33, "Spceified to limit the usage of gpu")
Even reducing layers from 9 to 5 and batch limiting too