Why is it not feasible to reduce the number of convolution channels from high to low (128 β 64)? <img width="815" height="846" alt="Image" src="https://github.com/user-attachments/assets/ecce9d6f-7fed-4dce-9640-36b519733a58" />
Why is it not feasible to reduce the number of convolution channels from high to low (128 β 64)?