NCS for self driving RC car

[idata]

I am currently working on a self driving RC car on my university and the raspberry pi performance is not enough so i decided to get NCS and use it with the PI but my teammate told me that NCS only supports a single input variable. Good for image classification but not vehicles. i would like to validate this statement,

Thanks,

[idata]

@GoldenWings Hi, can you explain what you mean regarding to the "NCS only supports a single input variable" and what kinds of tasks you have in mind for the NCS usage in terms of your project? The NCS is great for image classification as you mentioned and I'm excited that you are considering the NCS for your autonomous RC car project, but I would like to know more details so that I can better answer your question. Thanks!

[idata]

@Tome_at_Intel Hi, my expectation from NCS is, i wish to plug to raspberry pi 3 model b and compile a neural network that detects lanes and obstacles in the track we will start of by deploying PilotNet which is a CNN that takes in image input and outputs steering commands then we might extend the neural network to another architecture which takes in sensor data as input including camera and output a steering commands.

Another question in mind am i able to deploy some kind of Modular neural network architecture on NCS?

NOTE: we are planning to use Tensorflow on the Raspberry PI 3.

[idata]

@GoldenWings You have a rather interesting project going on. Although the NCS does support the Raspberry Pi 3 and Tensorflow, multiple inputs is a feature that we are still looking into and at the moment we don't currently have support for it.

[idata]

@Tome_at_Intel regarding multiple inputs is it something related to the sdk or to the hardware it self?

Another thing do you NCS is a good choice for tasks i mentioned.

[idata]

hi

As we got to know that now we can use tensorflow created model in NCS. As in there document they have specify that for custom created model we will need graph, stat.txt and categories.txt file in which categories.txt and stat.txt file is ok. But my question is that as we know that while training on tensorflow we get model.ckpt.meta, model.ckpt.index , model.ckpt.data and greaph.pb file. so when i am giving this graph.pb file in NCS api . its giving me error. So from how we can create graph file without weight (not graph.pb)? what is the difference between this graph.pb file and the graph file used in NCS.

[idata]

@GoldenWings The multiple inputs issue is related to the SDK and not the hardware. Multiple inputs is a feature that we have planned for a future release. The NCS is does well with CNNs and image related networks, but I can't say for sure how well it would fit into your project.

[idata]

@Tome_at_Intel any idea when multiple inputs might be released?

[idata]

@GoldenWings While multiple inputs is a feature that we are looking into, I cannot provide a road map or ETA for this feature. I do want to let you know that we listen to and consider all developer requests and try to incorporate that into our SDK for future releases.

[idata]

@imran , The graph.pb you mentioned, you made it from export_inference_graph.py from tensorflow's object detection api right ? And If yes, then that won't work on NCS. You'll have to compile the graph using mvNCCompile file provided with the NCS SDK. The graph file generated by mvNCCompile is used by the NCS.

[idata]

@akash.jobanputra ,

This i know, that is what our question is , How to create that graph file using mvNCCompile for ssd_mobilnet model.and this is the problem because different in file format of inception model and ssd_mobilnet model

[idata]

@Imran why not create new discussion to keep this discussion on topic?

[idata]

@akash.jobanputra , have you implemented it by yourself? Had you implemented it you might have got the discussion we are having it here. Its not about what how to create a graph file from .meta file but the issue is about the issue faced while doing it. If you have done it , just give the solution, instead of beating around the bushes, that would help.