HI, I am doing my next-to-last project of the 3 term Self driving Car program @ Udacity. It entails modifying the VGG16 model to make it a Fully Convolutional Network… After training my network on a work-machine with NVidia GPU, I wanted to use the Movidius NCSDK I purchased to run the test pictures through… At this point, I have the NCSDK up and running, I saved the trained model (generated .index, .meta and .data-00000-of-00001 files). My next steps are why I reaching out to you for help.
1) do I need to generated a .pb file or the .index/.mata/… files I generated are useful?
2) Do you have sample code showing usage of VGG16?
3) any guidance you can provide would be much appreciated as I would like to keep using the NCSDK stick as I expand my knowledge
@KahrenA We have added VGG16 support in the latest release.
Hi, i have realized a model VGG16, download from trained model link
when use a *.pb file -> result ok
Compile a model -> no error
when use a *.graph file -> result wrong
to run for example:
python3 prediction.py --configfile ./config_cat_inception.json --model ./cat_inception.pb --test ./test_images.jpeg
python3 prediction.py --configfile ./config_cat_inception.json --model ./cat_inception.graph --test ./test_images.jpeg
use dogs or cats images.
Thanks in advance