- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello all,
I am experimenting with the application Video_face_matcher_multiFace which uses Facenet.
I made some changes in the code which reads images. And I'm having the following error:
Traceback (most recent call last):
File "./matcher.py", line 349, in
sys.exit(main())
File "./matcher.py", line 322, in main
out, _ = run_inference(validated_image, graph)
File "./matcher.py", line 57, in run_inference
out, userobj = get_graph_result(facenet_graph, im.astype(numpy.float16))
File "./matcher.py", line 34, in get_graph_result
graph.LoadTensor(img, None)
File "/usr/local/lib/python3.5/dist-packages/mvnc/mvncapi.py", line 253, in LoadTensor
raise Exception(Status(status))
Exception: mvncStatus.ERROR
Graph recompilation didn't help. Any ideas what could have caused this error?
Here is the part of main() which I changed:
def main():
use_camera = True
# Get a list of ALL the sticks that are plugged in
# we need at least one
devices = mvnc.EnumerateDevices()
if len(devices) == 0:
print('No NCS devices found')
quit()
# Pick the first stick to run the network
device = mvnc.Device(devices[0])
# Open the NCS
device.OpenDevice()
# The graph file that was created with the ncsdk compiler
graph_file_name = GRAPH_FILENAME
# read in the graph file to memory buffer
with open(graph_file_name, mode='rb') as f:
graph_in_memory = f.read()
# create the NCAPI graph instance from the memory buffer containing the graph file.
graph = device.AllocateGraph(graph_in_memory)
face_vectors = []
for person in white_list:
person_imgs = os.listdir(os.path.join('./validated_images/', person))
person_vectors = []
tmp = []
for i in person_imgs:
validated_image = cv2.imread(os.path.join("./validated_images/", person, i))
out, _ = run_inference(validated_image, graph)
if(len(out) != 0):
person_vectors.append(out)
tmp.append(out) # numpy.ndarray.flatten(out))
tmp = numpy.array(tmp).astype('float32')
# Use k-means to separate data into clusters:
k = round(max(1, len(tmp) / 4))
compactness, labels, centers = cv2.kmeans(tmp, k, None, (cv2.TERM_CRITERIA_COUNT|cv2.TERM_CRITERIA_EPS, 1000, 0.001),
10, cv2.KMEANS_PP_CENTERS)
face_vec = get_kmeans_clusters(person_vectors, labels, len(centers))
face_vectors.append(face_vec)
#face_vector = numpy.zeros(valid_output[0].shape)
In fact, graph can load without a problem any two tensors and then it fails to load any third tensor, no matter which pictures I give it.
Thank you a lot in advance
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@jelena It seems like there isn't a img tensor based on your log (below). graph_get_result has im, but img is being passed to LoadTensor()? I don't see your entire code so I am just assuming this is the error.
out, userobj = get_graph_result(facenet_graph, im.astype(numpy.float16))
File "./matcher.py", line 34, in get_graph_result
graph.LoadTensor(img, None)
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@Tome_at_Intel thanks for your response,
the image is being passed to LoadTensor like this:
def get_graph_result(graph, img):
graph.LoadTensor(img, None)
print("Loaded")
return graph.GetResult()
The function get_graph_result is being called in a loop and any 2 images (=> 2 first iterations) are processed successfully, and any 3rd image causes this error, no matter which images I put in my data folder.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@jelena Please make sure you are running the preprocessing steps on every image you read in like the original app.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@Tome_at_Intel thank you. I apologize for not responding for so long. I do all the preprocessing steps just like the original app:
//Detect a face and crop the rectangle:
108 def get_face_rect(image):
109 detector = dlib.get_frontal_face_detector()
110 dets = detector(image, 1)
111 roi_color = []
112 faces=[]
113
114 for d in dets:
115 cv2.rectangle(image,(d.left(),d.top()),(d.right(),d.bottom()),(255,0
116 # roi_gray = gray[y:y+h, x:x+w]
117 roi = image[d.top():d.bottom(), d.left():d.right()]
118 faces.append((d.left(), d.top(), d.right()-d.left(), d.bottom()-d.to
119 roi_color.append(roi)
120 return roi_color, faces
121
//Do the preprocessing:
125 def preprocess_image(src):
126 # scale the image
127 NETWORK_WIDTH = 160
128 NETWORK_HEIGHT = 160
129 preprocessed_images, face_rect = get_face_rect(src)
130
131 for im in preprocessed_images:
132 im = cv2.resize(im, (NETWORK_WIDTH, NETWORK_HEIGHT))
133
134 #convert to RGB
135 im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
136
137 #whiten
138 im = whiten_image(im)
139 # return the preprocessed image
140 return preprocessed_images, face_rect
141
33 def get_graph_result(graph, img):
34 graph.LoadTensor(img, None)
35 print("Loaded")
36 return graph.GetResult()
37
44 def run_inference(image_to_classify, facenet_graph):
45
46 # get a resized version of the image that is the dimensions
47 # SSD Mobile net expects
48 resized_images, face_rect = preprocess_image(image_to_classify)
49
50 # ___________________________
51 # Send the image to the NCS
52 # ___________________________
53 output = []
54 print("size = ", len(resized_images))
55 for im in resized_images:
56 cv2.imshow('',im)
57 cv2.waitKey(0)
58 out, userobj = get_graph_result(facenet_graph, im.astype(numpy.float
59 print('OK')
60 output.append(out)
61 return output, face_rect
62
Thank you in advance for any help!
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@jelena I see that you are running a face detection (detect_face_rect()) on the source frame, cropping and saving the detected faces to preprocessed_faces, and then trying to perform preprocessing on the detected faces in your for loop. You aren't actually doing any preprocessing because in lines 132-138, you are assigning the preprocessed data to the iterator variable (im) and then returning the preprocessed_faces array.
Keep in mind that the NCS can only perform inference on one image at a time. If you are using NCSDK v 2.xx.xx you can queue up inferences and read them in a FIFO manner.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
@Tome_at_Intel thank you, you are right! I was absolutely sure that such cycle created a reference, not a copy of an element in Python…
Thanks for the note about images inference, too
![](/skins/images/7FC17B7B85029576C25F1E43CE255B51/responsive_peak/images/icon_anonymous_message.png)
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page