You are on page 1of 2

COMPUTER VISION

HW2

RAVINDRA SAMRIA

22565013

Ans1(i)

Input channel- 512

Dim – 1024*1024

After applying 1*1 convolution layer with 128 filter

Output spatial dim- 1024*1024

With channel-128

(ii)

2*2 pooling with stride 2

Output spatial dimension-512*512 with channel -512

Ans 2 .

Parameter will be 9*9*512

Ans 3.

Both input and output spatial dimension are same it means 1*1 convolution with channel – 512

Ans 4.

(i)

TOAL MAC -1024*1024*7*7*512*64

(ii)

1024*1024*1*1*512*32

And for 64 filter and 7*7 layer

1024*1024*7*7*32*64
Ans 5.

(I)

FOR INCEPTION

PARAMETERS- 388096 FLOPS- 304267264

FOR 3*3 LAYER

PARAMETER- 1105920 FLOPS- 867041280

FOR 5*5 LAYER

PARAMETER- 3072000 FLOPS- 240848000

(II)

PATHS PARAMETERS FLOPS


1 32768 25690112
2 253952 199098368
3 84992 66633728
4 16384 12845056

Ans 6.

(i)

No of groups = 4 for group convolution

(ii)

In standard convolution

Input channel – 8

Output channel = 4 (filters)

You might also like