model got low mAP

[akhilgkrishnan]

Hey @AlexeyAB ,

I trained a model for detecting alcohol scenes in videos. So I labeled the image of Alcohol bottles and alcohol glass in a single class. I got only 18% of mAP @0.5.
I used the yolov3 model and trained in google colab.
For my dataset, it contains a lot of images having multiple alcohol bottle objects in a single image, but I labeled only a few of them. Is that is the problem for low mAP..?
How to solve this..?
Hey @AlexeyAB

[readicculus]

If a label is in the image you want it labeled, otherwise when the model is training it penalize what should be a correct detection. Doesnt have to be all of them but definitely want most examples labeled.

[lsd1994]

Label all you want to detect, otherwise it's considerated to be background.

[akhilgkrishnan]

Hey @readicculus @lsd1994 @AlexeyAB ,
I labeled the beer bottle and beer glass in one class called "Alcohol". When I trained this model I got only 20 mAP. Is that is the problem? Do I need to label within a separate class ..?

Thanks

[lsd1994]

Please show some images with label.

[akhilgkrishnan]

Hey @lsd1994 ,
sample label image

[lsd1994]

It's hard to say, maybe your testset is not similar to trainset.

[akhilgkrishnan]

@lsd1994, my dataset drive link shared below, please check my dataset
https://drive.google.com/file/d/1lCi12xMHdXwWlxIAAyPSfKpyvJyl6AcD/view?usp=sharing