I have an image, with many fruits on RGB image. I don't have ground truth. my interest is, to count how many fruits on image using Deep Learning, which is the best way to count them. Is there any best reference?
Try MobileNet SSD + dnn(cv2.dnn) module in OpenCV to build your object detector.
SSD is very simple and faster than R-CNN and some time more accurate than YOLO.
Following link might be helpful for you:
Article Deep Count: Fruit Counting Based on Deep Simulated Learning
https://www.mathworks.com/matlabcentral/answers/321355-issue-in-recognising-multiple-objects-in-an-image
http://ijarcet.org/wp-content/uploads/IJARCET-VOL-3-ISSUE-6-2164-2167.pdf
Hi Rami,
I have not done fruit counting myself but I have worked on a similar topic for cell counting in histologist images. You need to perform instance segmentation first and then get the number of detected object in the image.
For instance segmentation they are many implementations available but as far as I know they are mostly based on supervised deep learning-based approaches. One of the most well-known one is "Maks-RCNN" and you can find the source code from here:
https://github.com/matterport/Mask_RCNN
If you objects (fruits) are not overlapped you could also use simpler approaches such as "U-Net" or "fully convolutional neural networks".
Keras implementation of U-net:
https://github.com/jocicmarko/ultrasound-nerve-segmentation
Hope it helps.
Amirreza
- Badges
- Science topic
- Similar topics
- Computer Science
- Images