Amount of custom filters in CNNs when comparing to default filters?

When I want to compare the performance of CNNs between using the default and using custom filters, should I use the same amount of filters in both cases?
Let’s say I want to use custom filters (and by that I mean I choose my own initial weights for the filters) for the first conv layer. Should I create the same amount of custom filters as my “normal” CNN has in its first layer?

Hey there @undefined
When comparing CNNs with default and custom filters, it’s important to use the same number of filters in both cases. This ensures a fair comparison. Keeping the number of filters consistent allows us to see how customizing filters affects performance without other variables getting in the way. It also ensures fair use of computational resources and training conditions. If you’re curious about how different filter numbers impact performance, it’s best to conduct separate experiments while keeping other factors constant.