-1

I am looking at object detection using CNN, I understand there are several models such as a sliding window, Region-based CNN, Fast Region based and finally Yolo. I wish to implement Yolo model in Tensor flow and have searched on net but could not find something helpful. Can anybody comment how to implement Yolo model using Tensor flow and on the same note I understand that there is a Tensorflow API for object detection, whats the difference between using this API and directly applying Yolo model? Any online material that could be helpful to write the code for this purpose. regards

Nhqazi
  • 732
  • 3
  • 12
  • 30

2 Answers2

1

You are looking for https://github.com/thtrieu/darkflow, which is yolo implementation in tensorflow. Check our "Siraj raval" video here on the topic. https://www.youtube.com/watch?v=4eIBisqx9_g&vl=en

You can use tensorflow object detection or Yolo , they are two different implementations on object detection. In case the object that you are interested in is not part of the pre-trained objects list, you can use transfer learning for re-training. Hope this helps!

Srinivas Bringu
  • 452
  • 3
  • 9
  • Thanks Srinivas does this mean that Tesnorflow object detection does not follow Yolo model.? Can u please pass me some links on it. I assume you are referring to Tensor flow object detection API. ??? whats different between using yolo and Tensor flow detection model? – Nhqazi Aug 26 '18 at 08:25
1

Moving the response to answer, as its too long for a comment:

Tensorflow is a platform and there is a way to implement object detection, refer this SO post for examples, its tensorflow object detection Tensorflow real time object detection There are several tensorflow based standard models like SSD, RCNN etc., which do object detection. Yolo is seperate idea/implementation which was originally not built on tensorflow, but has also been adopted on to tensorflow and is called darkflow, which is what i shared in previous response. The primary difference between Yolo and standard tensorflow based models is the internal approach of how image is divided into grids, regions are proposed for detected and rest of the pipeline. You can read more on web for more details of individual approaches

Srinivas Bringu
  • 452
  • 3
  • 9