OPs TableΒΆ

IOPs(Interger operators) = Op Weight * Output Shape Size

The OPs of deterministic operator is as below:

Op Name Op Weight
variable 1
conv2d, conv2d_transpose (without bias) $w_2w_3w_4*3$
conv2d, conv2d_transpose (with bias) $w_2w_3w_4*3+1$
dense (without bias) $w_{1}*3$
dense (with bias) $(w_{1}+1)*3$
non_max_suppression $input_{0,2}*20$
max_pool2d $poolsize^2$
__add_scalar__ 1
__add_symbol__ 1
__div_scalar__ 1
__div_symbol__ 1
__equal_symbol__ 1
__greater_equal_symbol__ 1
__greater_symbol__ 1
__left_shift_symbol__ 1
__less_equal_symbol__ 1
__less_symbol__ 1
__lshift_scalar__ 1
__max_symbol__ 1
__min_symbol__ 1
__mod_symbol__ 1
__mul_scalar__ 1
__mul_symbol__ 1
__not_equal_symbol__ 1
__pow_scalar__ 1
__rdiv_scalar__ 1
__right_shift_symbol__ 1
__rshift_scalar__ 1
__rsub_scalar__ 1
__sub_scalar__ 1
__sub_symbol__ 1
__undef__ 1
abs 1
add 1
argmax 1
argmin 1
broadcast_add 1
broadcast_div 1
broadcast_equal 1
broadcast_greater 1
broadcast_greater_equal 1
broadcast_left_shift 1
broadcast_less 1
broadcast_less_equal 1
broadcast_max 1
broadcast_min 1
broadcast_mod 1
broadcast_mul 1
broadcast_not_equal 1
broadcast_right_shift 1
broadcast_sub 1
broadcast_to 1
cast 1
clip --2
collapse_sum 1
concatenate 1
copy 1
cvm_clip 2
cvm_left_shift 1
cvm_lut 10
cvm_right_shift 1
div 1
elemwise_add 1
elemwise_div 1
elemwise_mod 1
elemwise_mul 1
elemwise_sub 1
elemwise_sum 1
expand_dims 1
expand_like 1
flatten 1
flip 1
full 1
full_like 1
gather_nd 1
get_valid_counts 1
global_max_pool2d 1
greater 1
less 1
log 1
log2 1 (constant time ~64)
logical_and 1
logical_not 1
logical_or 1
matmul 1
max 1
mean 1
min 1
multiply 1
negative 1
nn.relu 1
ones 1
ones_like 1
prod 1
relu 1
repeat 1
reshape 1
reshape_like 1
sigmoid 1
slice 1
slice_like 5
split 1
sqrt 1
squeeze 1
strided_slice 5
subtract 1
sum $x_3*x_4$
take 10
tanh 1
tile 5
transpose 5
upsampling 1
where 1
zeros 1
zeros_like 1