Tensorboard观察graph 保存sess.graph 读取graph进行参数量计算量统计

tensorflow参数量、运算量统计

def stats_graph(graph):
    flops = tf.profiler.profile(graph, options=tf.profiler.ProfileOptionBuilder.float_operation())
    params = tf.profiler.profile(graph, options=tf.profiler.ProfileOptionBuilder.trainable_variables_parameter())

    print('FLOPs: {};    Trainable params: {}'.format(flops.total_float_ops, params.total_parameters))

保存graph(tensorboard观察) sess.graph

summary_writer = tf.compat.v1.summary.FileWriter('/home/allen/Documents/lck/Resnet-FPGA/DoReFa-Net/log/', graph)

$ tensorboard --logdir log

saver.save(sess) 或者 saved_model.simple_save()

####### saver.save
 saver = tf.compat.v1.train.Saver()
 saver.save(sess , '/home/allen/Documents/lck/Resnet-FPGA/DoReFa-Net/log/model_new')
 ###### savemodel
inp = input_tensors[0]  ## input is needed
outp = output_tensors[0]
tf.compat.v1.saved_model.simple_save(sess ,'/home/allen/Documents/lck/Resnet-FPGA/DoReFa-Net/logs',inputs={"myInput": inp},
             outputs={"myOutput": outp})