Yes, object size can affect image recognition accuracy, as models might struggle to detect very small or very large objects in an image. Small objects can be missed if the resolution is insufficient or if they lack distinguishing features, while large objects may require additional scaling or preprocessing.
Techniques like multi-scale object detection, used in models like Faster R-CNN or YOLO, address this challenge by analyzing images at different resolutions to capture objects of varying sizes. Data augmentation methods, such as resizing or cropping, also help models generalize across different object scales.
Ensuring balanced representation of object sizes in training datasets and using models optimized for scale invariance can significantly improve recognition accuracy in real-world scenarios.