- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
hi ,
i'm trying to run a script with NCS2 stick .i converted facenet model using model optimiser with datatype FP16 and while executing i should get array of embedding with size (1,512) but i got an array with of -1 with shape (1,512) ,could u help me how to resolve that error.
openvino version (2019.1.144)
ubuntu version(16.04)
Thanks
bindu sri.
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Dear donthireddy, bindu,
I don't understand your error message. Did you get a specific error or exception from Inference Engine ?
Thanks,
Shubha
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
hi shubha ,
if u go through the script in that i am using facenet.xml to extract 512 embeddings of an image but i got this array exec_net============= {'normalize': array([[ 1., -1., -1., 1., -1., 1., -1., 1., -1., -1., 1., 1., -1.,
-1., -1., -1., 1., -1., 1., -1., -1., 1., 1., -1., -1., -1.,
-1., 1., 1., 1., -1., 1., 1., -1., -1., -1., -1., -1., -1.,
-1., -1., 1., 1., -1., -1., 1., -1., -1., -1., -1., 1., 1.,
1., -1., -1., 1., -1., 1., 1., 1., -1., 1., -1., -1., -1.,
1., 1., -1., 1., -1., 1., 1., -1., 1., -1., -1., -1., -1.,
-1., 1., 1., -1., -1., -1., 1., -1., -1., 1., -1., 1., 1.,
-1., 1., -1., 1., 1., 1., -1., -1., 1., 1., -1., -1., 1.,
1., 1., 1., -1., -1., -1., 1., -1., -1., -1., -1., -1., 1.,
-1., 1., 1., -1., 1., -1., -1., -1., -1., 1., -1., -1., -1.,
-1., -1., 1., -1., -1., 1., -1., 1., 1., 1., -1., -1., 1.,
-1., 1., -1., -1., -1., -1., -1., 1., -1., -1., 1., 1., -1.,
1., -1., 1., -1., 1., -1., -1., -1., 1., -1., 1., -1., -1.,
1., -1., 1., -1., -1., 1., -1., 1., -1., -1., -1., -1., -1.,
-1., 1., -1., 1., -1., 1., -1., 1., 1., -1., 1., -1., 1.,
-1., 1., -1., -1., -1., 1., 1., -1., -1., -1., -1., -1., -1.,
-1., 1., -1., -1., -1., -1., -1., -1., -1., 1., -1., 1., -1.,
-1., -1., -1., -1., -1., -1., -1., -1., 1., 1., -1., -1., -1.,
-1., -1., 1., -1., -1., -1., 1., 1., 1., -1., 1., 1., -1.,
1., -1., -1., -1., 1., 1., 1., -1., -1., 1., -1., 1., 1.,
-1., -1., 1., -1., -1., -1., -1., 1., -1., -1., -1., 1., 1.,
-1., -1., -1., -1., -1., -1., -1., -1., 1., -1., -1., -1., -1.,
-1., 1., -1., -1., -1., -1., -1., -1., -1., -1., 1., 1., -1.,
-1., 1., -1., 1., -1., -1., -1., -1., -1., 1., -1., -1., -1.,
-1., 1., 1., -1., -1., -1., -1., -1., 1., -1., -1., 1., -1.,
-1., 1., -1., 1., 1., 1., 1., -1., 1., 1., 1., -1., -1.,
1., 1., 1., -1., 1., -1., -1., -1., 1., -1., 1., 1., -1.,
-1., -1., 1., -1., -1., 1., -1., -1., -1., -1., -1., -1., -1.,
-1., 1., 1., -1., -1., 1., -1., -1., -1., 1., 1., 1., -1.,
-1., -1., -1., -1., -1., -1., 1., 1., -1., 1., -1., -1., -1.,
1., 1., -1., -1., -1., 1., 1., -1., 1., 1., -1., 1., -1.,
1., -1., -1., -1., -1., 1., -1., 1., -1., 1., -1., -1., -1.,
1., 1., -1., -1., 1., 1., -1., 1., -1., -1., 1., -1., -1.,
-1., 1., 1., -1., -1., -1., 1., 1., -1., -1., -1., -1., -1.,
1., -1., -1., -1., -1., 1., -1., -1., -1., -1., 1., -1., -1.,
-1., 1., 1., 1., 1., -1., -1., -1., -1., 1., -1., -1., -1.,
-1., -1., 1., -1., 1., -1., -1., -1., 1., -1., 1., 1., 1.,
1., 1., -1., -1., -1., -1., 1., -1., -1., -1., -1., -1., -1.,
1., -1., 1., -1., -1., -1., -1., 1., -1., 1., -1., -1., -1.,
1., -1., -1., -1., -1.]], dtype=float32)}
so i would like to know how ill get embeddings with that .xml
thanks
bindusri.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Dear donthireddy, bindu,
I studied your code and there is a lot of pre-processing going on. For instance, this function:
def prewhiten(x): mean = np.mean(x) std = np.std(x) std_adj = np.maximum(std, 1.0/np.sqrt(x.size)) y = np.multiply(np.subtract(x, mean), 1/std_adj) return y
Not to mention cropping and so on.
As I mentioned to this poster with a similar question and similar code also the reason you're getting these garbage values is because you are not properly telling model optimizer about the pre-processing of your images. Once you do that, you should not get these garbage values.
Hope it helps,
Thanks,
Shubha
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
hello shubha,
As you said i removed all the preprocesing functions and just executed the following code i got the same result.could just go through it and let me know the changes.
img ='deepu.jpg'
image = cv2.imread(img)
face_path = '/home/icsltd/Desktop/deepika/facenet-master/facenet-master/facenet.xml'
weights_file = face_path[:face_path.rfind('.')] + '.bin'
net = ie.IENetwork(face_path, weights_file)
plugin = ie.IEPlugin("MYRIAD")
input_name = list(net.inputs.keys())[0]
output_name = list(net.outputs.keys())[0]
exec_net = plugin.load(net)
print('first__________________________',image.shape)
n, c, h, w = net.inputs[input_name].shape
print('before__________________________',image.shape)
if image.shape[:-1] != (h, w):
print('__________________________',image.shape)
image = np.resize(image,(n,c,w, h))
print(image.shape)
# Transpose image for channel first format
#image = image.transpose([0,1, 3, 2])
feed_dict = {input_name: image}
print('++++++++++++++++++++++++====',feed_dict)
emb_outputs=exec_net.infer(feed_dict)
print('exec_net=============',emb_outputs)
thanks,
bindu sri.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Dear donthireddy, bindu,
Please download 2019R2 which was just released last week and try again.
Thanks,
Shubha

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page