cnn_full_model_epoch_42.data-00000-of-00001 (for yolov3) should be download by the baiduyun, for detail you can open the cnn_full_model_epoch_42.data-00000-of-00001 file.

Before training the model you defined, you'd better make sure the type and shape of the input, also the output of the model, as in ckpt2pbtxt.py,

	input_x = tf.placeholder(tf.float32, shape=[None, shapes_w, shapes_h, shapes_c], name=input_node)
	logits = modelOutput(input_x, class_num)

the line above requests to be changed according to your model, for more information please check in ckpt2pbtxt.py. Make sure the related ckpt files for the trained model are saved correctly and the related changes are made in ckpt2pbtxt.py. Then , run the follow cmd:

	python ckpt2pbtxt.py

1. Obtain the input and output node of the model, check in the xxx.pbtxt. Usually, the ouput node is the last node before the node named startwith save/..., like:

	node{
	
		name: "comExp/output/Conv2D"  
		
		#comExp is the defined variables_scope, output is the name defined for output layer, 
		
		#Conv2d is the corresponding op.
		
		op: "Conv2D"
		
		input: "comExp/MaxPool_2"
		
	... ...

2. Based the name of output node in xxx.pbtxt, run the follow cmd to finish the turn op:

	python pbtext2pb.py

3. To visual or test the .pb files, run the follow cmds:

	python tensorboardOfPb.py

make sure the built .pb files can be tested successfully.

And with the nodeOfPb.py, you can check the node in the generated result.txt. The node also can be obtain by running the follow cmd based on the ckpt files:

	python nodeOfCkpt.py

1. Run the follow cmd:

	./tensorflow2ncnn xxx.pb ncnn.param ncnn.bin

2. Check the model defined in ncnn.param, the model name and other information can be read in ncnn.param, the input and output name will be used for build the model in ncnn.

For more information of pbtxt2pb.py, please check: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/tools/freeze_graph.py