Thank you for the insight Joseph, Satya Malik and Phil Nelson. Please schedule this kind of review often between Roboflow and others (OpenCV, Luxonis OAK, etc). Request for next episode: YOLO, Scaled YOLO, YOLOR and TensorRT.
you guys encourage every thing but edge devices are mainly focused on tensorflow the pattern tensorflow is growing is like arduino they re providing hardware and software just like android tensorflow could take over because of their versatility in edge devices and there are more reasons to switch over to tensorflow soon yolo could be abandoned
Hi! Awesome video! I want to begin by saying I am new to machine learning and still learning. I have a couple of questions regarding what I am trying to accomplish using machine learning. I am trying to segment eyeglasses from selfies, ie. extract their shape. The selfies will be taken from the front. Ideally, I would want to segment one, or both lenses from the glasses, and if this is not possible, the whole piece of eyewear. Is this possible to do by manually label/annotate pictures, and later use transfer learning with a model trained on coco dataset? If so, how many pictures would I approx. need to annotate? Do you have any other better idea?
@@Vocal4Local thanks for the reply! I started working on mask r CNN but confusion about the annotation formats. Which is better. Detection or mask r cnn
@@saad2115 Based on my experience, there is no way to say which is better. You need to make a model with the same data then test for yourself to see which is better.
Thank you for the insight Joseph, Satya Malik and Phil Nelson.
Please schedule this kind of review often between Roboflow and others (OpenCV, Luxonis OAK, etc).
Request for next episode: YOLO, Scaled YOLO, YOLOR and TensorRT.
This video helped me a lot to get a overview of the yolo object detection family. Thank you for that.
30:10 (Bookmark)
Thank you
awesome!
you guys encourage every thing but edge devices are mainly focused on tensorflow the pattern tensorflow is growing is like arduino they re providing hardware and software just like android tensorflow could take over because of their versatility in edge devices and there are more reasons to switch over to tensorflow soon yolo could be abandoned
Hi! Awesome video!
I want to begin by saying I am new to machine learning and still learning.
I have a couple of questions regarding what I am trying to accomplish using machine learning.
I am trying to segment eyeglasses from selfies, ie. extract their shape. The selfies will be taken from the front.
Ideally, I would want to segment one, or both lenses from the glasses, and if this is not possible, the whole piece of eyewear.
Is this possible to do by manually label/annotate pictures, and later use transfer learning with a model trained on coco dataset?
If so, how many pictures would I approx. need to annotate?
Do you have any other better idea?
You'll need something called semantic segmentation. There's U-Net and ENet for this, and you could also use Detectron or Mask-RCNN
@@Vocal4Local thanks for the reply! I started working on mask r CNN but confusion about the annotation formats. Which is better. Detection or mask r cnn
@@saad2115 Based on my experience, there is no way to say which is better. You need to make a model with the same data then test for yourself to see which is better.
This is a marketing video, not a webinar. Talked too much, said too little. :(