I have two cameras to capture images for training keypoint detection and semantic segmentation model. One camera has smaller field of view and the other has larger field of view. Let's say, I capture two sets of data using both the cameras and train the model on each set of data. Will the two models have same performance? If not, how the field of view affects the performance of the model?
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
I think the you will see a difference in performance on two levels.
Difference in understanding of image: What I mean by this point is how the model perceives the image. Reason being presence of more information in each image as the FOV increases. I wonder if the chances of having multiple instances also increases in your case as the FOV increases. For example: Take an example of fish in culture tanks(you can imagine other things as well like people on zebra crossing), this would drastically impact the performance as the task is more difficult now.
Performance as measured my metric: Again if the FOV is larger the relative size of objects will be different for both cameras. This means you will have to keep in mind to scale parameters for metrics like OKS as FOV changes.
I would like to see the differences you found in your experiments.
