Training with partially annotated images #7797
-
|
I have dataset of 15 classes and some images are duplicated with different class names to annotate separately. Example, Img1.jpg - - - has a person and car but only car is annotated img1.jpg (renamed-same image as above ) - - - only car is annoated My question how does this effect the training, does yolo consider background as negatives? What's the easy way around this solution, combine all annotations per image? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
Your question is: 2 identical images, but only the car is annotated in the first image, and only the car (again?) is annotated in the duplicate 2nd image. Did you make a mistake? If you mean that you only annotate the car in the first image, and only the person in the 2nd image, it will not work. You'll destroy the training. See https://www.ccoderun.ca/programming/darknet_faq/#image_markup and https://www.ccoderun.ca/darkmark/ImageMarkup.html |
Beta Was this translation helpful? Give feedback.
-
|
Thanks 👍 Is there an easy way to know during the training that the network is overfitting. |
Beta Was this translation helpful? Give feedback.
Your question is: 2 identical images, but only the car is annotated in the first image, and only the car (again?) is annotated in the duplicate 2nd image. Did you make a mistake?
If you mean that you only annotate the car in the first image, and only the person in the 2nd image, it will not work. You'll destroy the training.
See https://www.ccoderun.ca/programming/darknet_faq/#image_markup and https://www.ccoderun.ca/darkmark/ImageMarkup.html