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.