High Performance Implementation of 3D Convolutional Neural Networks on a GPU

Table 2

Convolution layers of a 3D network; the filter size in all layers is , and the GFLOPS columns calculate the number of flops operations in each convolutional layer. Assume the batch size is .

Layer	× × × ×		GFLOPS
conv1	3 × 16 × 112 × 112 × 32	32	16.65
conv2	32 × 16 × 56 × 56 × 32	64	88.8
conv3	64 × 8 × 28 × 28 × 32	256	88.8
conv4	256 × 4 × 14 × 14 × 32	256	44.4
conv5	256 × 2 × 7 × 7 × 32	256	5.55