1
$\begingroup$

I have roughly 30,000 images of two categories, which are 'crops' and 'weeds.' An example of what I have can be found below:

enter image description here

The goal will use my training images to detect weeds among crops, given an orthomosaic GIS image of a given field. I guess you could say that I'm trying to detect certain objects in the field.

As I'm new to deep learning, how would one go about generating training labels for this task? Can I just label the entire photo as a 'weed' using some type of text file, or do I actually have to draw bounding boxes (around weeds) on each image that will be used for training? If so, is there an easier way than going through all 30,000 of my images?

I'm very new to this, so any specific details would really help a lot!

$\endgroup$

2 Answers 2

2
$\begingroup$

If each photo is intended to show a photo of weed or crops you should give one label. If your task is different where you also try to localize weed or crops in the image, then you need to label accordingly. My understanding is you are trying to do the first case, therefore, there should be one label for each image.

$\endgroup$
3
  • $\begingroup$ I am trying to pinpoint the location of the weeds in the field. Therefore, should I using something like the labelImg software to create Pascal VOC samples of the images? $\endgroup$
    – ihb
    Sep 15, 2020 at 1:59
  • 1
    $\begingroup$ I think so. The following link discusses about preprocessing training images for YOLO , you can follow the same. It will be really interesting to see how well your model works for this task. arunponnusamy.com/… $\endgroup$
    – gihan
    Sep 15, 2020 at 2:16
  • $\begingroup$ this is exactly what I needed. Thank you! $\endgroup$
    – ihb
    Sep 15, 2020 at 13:21
2
$\begingroup$

This is really a semantic segmentation problem if OP wants to pinpoint the weeds.

If OP wants to have such a segmentation he will need to hand segment every.single picture.

$\endgroup$
1
  • $\begingroup$ Maybe you should explicitly say that "semantic segmentation" is about classifying pixels (maybe provide an image that illustrates it) and consequently "segmenting" the objects in the image. Of course, labelling each pixel will be a lot of labor, so maybe you should also bring this up. $\endgroup$
    – nbro
    Sep 15, 2020 at 1:54

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .