search

yinguobing / cnn-facial-landmark

Training code for facial landmark detection based on deep convolutional neural network.
MIT License
628 stars 182 forks source link

same landmarks predictions for any image #88

Open vaibhavch opened 5 years ago

vaibhavch commented 5 years ago

Using butterfly module for interference. A custom dataset trained with 292 landmark points.

input node: butterfly/image_tensor

output_node: butterfly/layer6/final_dense

image = cv2.imread(args.image) face_img = np.array(image, np.uint8) marks = np.array(fly.run([face_img])).flatten()[:292] marks = np.reshape(marks, (-1, 2)) print(marks)

why iam getting same predictions for every image?

abdou31 commented 5 years ago

More explanation, could you show the results on images?

vaibhavch commented 5 years ago

@abdou31

Screenshot 2019-11-03 at 10 41 10 PM

same landmark coordinates irrespective of image input. something wrong with model itself?

abdou31 commented 5 years ago

I will help you, but did you can tell me where do you get the annotated dataset please

luoww1992 commented 4 years ago

I have the same result, have you solved it ? how to do it ?