I was reading the architecture of the YOLO v1 paper and was trying to tally the dimensions after each layer. I couldn't understand the "number of filters" dimension at the beginning two locations.
The input dimensions are 448x448x3, So after applying a convolutional layer 7x7x64, stride=2, padding=3, shouldn't the output be 224x224x64? And after max pooling using 2x2, stride=2, it would reduce to 112x112x64? But the paper says it is 112x112x192.
Could someone explain how it is 192?
