Dear Zhi
IN NeuronTree v3dneuron_GD_tracing 
Why is “trace to some selected markers” slower than “trace to the entire image”？
From my standpoint ,this method which trace to some selected markers has less endpoints than method that trace to the entire image，and  this method which trace to some selected markers has not pruning process

Several factors could affect the tracing speed. One of the most important factors is the background threshold. In traditional GD tracing, non-zero pixels are considered as the foreground. In APP1 tracing, the background threshold is defined by the user. There are some other factors, e.g. downsample option.

Dear Zhi
I have a idea . I want to add a level of confidence in APP.The criterion is to check if a large gap presents on an arbour by comparing the gap distance and a score determined by the mean radius sampled at the previous tracing steps.
But my programming level is poor,can you suggest me how to change this code of app1.

Thanks for interesting in improving the performance of our APP1 tracing. Unfortunately, we are not supporting in coding level. Please try to learn C/C++ by yourself and implement your idea on your own. There are lots of examples in our codebase. You should be able to catch up very quick.