[beginner's question]cudnn images and its feature maps

Accelerated Computing GPU-Accelerated Libraries
1

if anyone can tell me that in the cudnn, is every image has many feature maps, and if that, what that feature maps mean? in my conception, every image is sent to the net and then the net is trained by that, in the process, every layer has many feature maps. so what does the cudnnSetTensor4dDescriptor function’s parameter ‘n’ mean,which is described as number of images? aren’t that parameter only be 1?

2

i think parameter ‘n’ means the number of image u put into the network one times, also means the batchsize number