Technical blog and training materials

[idata]

Hello everyone,

We just launched a technical blog that will focus on building projects using Intel Movidius Neural Compute Stick and Neural Compute SDK. You can access the at https://movidius.github.io/blog/

Thanks

@AshwinVijayakumar

[idata]

I study a project on "raspberry pi 3" with "The neural compute stick" in order to identify the quantity dragon fruit on a garden. Could you please give me some material "The neural compute stick".

[idata]

Can you please put an example of object detection with the ncs?

[idata]

@monotosh, I'll look into creating a blog on object detection, until then you refer the MobileNet_SSD and tiny YOLO examples on https://GitHub.com/Movidius/ncappzoo

[idata]

@AshwinVijayakumar Any news about the blog on creating and running custom tensorflow on movidius ?

reference

[idata]

EDIT: We now have several tutorials on running TensorFlow models on NCS

1. On-demand webinar which highlights the entire workflow to run TF models on NCS - See 'Section 4 Part 1, timestamp 00:26:00) https://software.intel.com/en-us/videos/deploying-image-classifiers-on-intel-movidius-neural-compute-stick
2. Sample code and instructions to deploy pre-trained TF models
   • MobileNet - https://github.com/movidius/ncappzoo/tree/master/tensorflow/mobilenets
   • Inception - https://github.com/movidius/ncappzoo/tree/master/tensorflow/inception
   • InceptionResNet - https://github.com/movidius/ncappzoo/tree/master/tensorflow/inception_resnet_v2
3. Finetuned TF model with custom dataset - https://github.com/ashwinvijayakumar/ncappzoo/tree/flowers/apps/flowers
4. Custom network + custom dataset:
   • FashionMNIST - https://github.com/ashwinvijayakumar/ncappzoo/tree/fashion-mnist/tensorflow/fashion-mnist
   • Single convolution NN - https://github.com/ashwinvijayakumar/ncappzoo/tree/single-conv/tensorflow/single-conv

Tag @arshdeep

@GoldenWings, we ran into some bugs with respect to exporting TF models for inference, hence the delay. V1.12 has a fix, so I am curating the blog now. You can refer these examples until I have had a chance to polish & publish the blog.

~~MNIST example: https://github.com/ashwinvijayakumar/ncappzoo/tree/mnist/tensorflow/mnist~~

~~Fashion MNIST example: https://github.com/ashwinvijayakumar/ncappzoo/tree/fashion-mnist/tensorflow/fashion-mnist~~~~A super-simple custom network trained on fashion MNIST: https://github.com/ashwinvijayakumar/ncappzoo/tree/single-conv/tensorflow/single-conv~~

Developer guide explaining the changes needed to export a model for inference: https://movidius.github.io/ncsdk/tf_compile_guidance.html

[idata]

@AshwinVijayakumar: Is there any chance you can come up with an end-to-end solution from Keras -> movidius? Because I can not get this to work due to a myriad array of parser issues.

[idata]

@pisymbol, I have been looking into converting Keras models to TF and then to Movidius, but no joy. Until we get Keras support on NCSDK, I am afraid our only option is to rewrite Keras models in TensorFlow (I know, it's super tedious).

[idata]

@AshwinVijayakumar Can we make a custom image classifier with it ? Any example to do that? I am a newbee

[idata]

Please suggest

[idata]

@jaymj , not sure I understand the question. Are you asking if you can fine-tune a specific network with your own dataset? If so, here are some resources:

Caffe GoogLeNet dogs vs cats (same instructions can be used to train different networks with different datasets) - https://movidius.github.io/blog/deploying-custom-caffe-models/

Tensorflow MNIST trained on fashion-mnist dataset - https://github.com/ashwinvijayakumar/ncappzoo/tree/fashion-mnist/tensorflow/fashion-mnistCustom network using TF - https://github.com/ashwinvijayakumar/ncappzoo/tree/single-conv/tensorflow/single-conv

Recommended workflow for running TF models on NCS - https://github.com/ashwinvijayakumar/ncappzoo/tree/tensorflow-template/tensorflow/mobilenets. You can use export_inference_graph to export different networks and datasets.

As mentioned earlier, we don't yet have a proven method/workflow to run Keras models on NCS.