I need to train a model to count number of cylinders with a particular type of occlusion in the image.
Following is the image: https://i.stack.imgur.com/0VKJ7.jpg
As you can see in the image, the occlusion is of only one type - Top railing of truck
I do have only one image. Even with augmentation, I wont have much images to train.
How can I proceed?
I’d highly recommend collecting more images.
But if that is impossible you could try generating synthetic data.
hi Brad and Mohamed
thanks for your answers
i can create a synthetic dataset using the link but i guess what it will do is put my image in random images
rather, do u think is it a good idea if manually update the image and putting the top railing on top of cylinder in different positions and angles?
and i take up that manual task, how many images minimum do u think i need to make?
if i could take more images with different angle, daylight conditions
how many images would I need to take minimum?
Hi @Lakshay_Dulani, taking more images in variable conditions would be helpful. The best thing to do is get images that will be representative of the types of environments your model will be operating in (e.g indoor vs outdoor, dimly-lit area vs bright areas).
It’s hard to say the proper number of images, as it’s a function of variability. The uniqueness of the shape of the object, how differentiable it is in images from its surroundings, etc.
Your best bet is to start with a minimum of 50-100 annotated examples of the object, and then utilize Active Learning to more quickly improve model training and inference results.