diff options
author | Eric Kunze <eric.kunze@arm.com> | 2021-08-17 15:20:06 -0700 |
---|---|---|
committer | Eric Kunze <eric.kunze@arm.com> | 2021-08-23 16:26:33 -0700 |
commit | 8f57f9e092836a584d7b007b926a31fe2bc80b8a (patch) | |
tree | f2b8e7d21e022a3e089eae0d98e925707f90aaeb /chapters/tensor_ops.adoc | |
parent | ca2a854e3d46f91ecaa446d4b2311112cc2326fd (diff) | |
download | specification-8f57f9e092836a584d7b007b926a31fe2bc80b8a.tar.gz |
Fix pooling argument tables
Make the input tensor shape dimensions correct for the pseudocode.
Clarify the count used for average pooling only includes
valid elements.
Remove stale reference to PLACEHOLDER nodes.
Change-Id: Ia0b9f0aa404008c6a36671da12188cb0999712d4
Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Diffstat (limited to 'chapters/tensor_ops.adoc')
-rw-r--r-- | chapters/tensor_ops.adoc | 20 |
1 files changed, 12 insertions, 8 deletions
diff --git a/chapters/tensor_ops.adoc b/chapters/tensor_ops.adoc index 9a1c035..7334f67 100644 --- a/chapters/tensor_ops.adoc +++ b/chapters/tensor_ops.adoc @@ -77,7 +77,7 @@ This performs an average pooling over the given input tensor. A sliding window o |=== |Argument|Type|Name|Shape|Description -|Input|in_t*|input|[N,H,W,C]|Input tensor 4D +|Input|in_t*|input|[N,IH,IW,C]|Input tensor 4D |Attribute|int*|kernel|[2]|[kernel_y, kernel_x] |Attribute|int*|stride|[2]|[stride_y, stride_x] |Attribute|int*|pad|[4]|[pad_top, pad_bottom, pad_left, pad_right] @@ -109,9 +109,13 @@ for_each(0 <= n < N, 0 <= oy < H, 0 <= ox < W, 0 <= c < C ) { for_each(0 <= ky < kernel_y, 0 <= kx < kernel_x) { y = iy + ky; x = ix + kx; - acc_t value = tensor_read<in_t>(input, [N,IH,IW,IC], [n,y,x,c], input_zp, pad); + acc_t value = tensor_read<in_t>(input, [N,IH,IW,C], [n,y,x,c], input_zp, pad); acc = apply_add<acc_t>(acc, value); - if (0 <= y < IH and 0 <= x < IW) count++ + // Only values from the input tensor are used to calculate the + // average, padding does not count + if (0 <= y < IH and 0 <= x < IW) { + count++; + } } if (is_float(out_t)) { output_val = acc / (float)count; @@ -120,7 +124,7 @@ for_each(0 <= n < N, 0 <= oy < H, 0 <= ox < W, 0 <= c < C ) { acc = apply_scale_32(acc, scale.multiplier, scale.shift, false); output_val = (in_t)apply_clip<acc_t>(acc + output_zp, minimum<in_t>, maximum<in_t>) } - tensor_write<in_t>(output, [N,H,W,OC], [n,oy,ox,oc], output_val); + tensor_write<in_t>(output, [N,H,W,C], [n,oy,ox,c], output_val); } ---- @@ -428,7 +432,7 @@ This performs a max pooling over the given input tensor. A sliding window of siz |=== |Argument|Type|Name|Shape|Description -|Input|in_t*|input|[N,H,W,C]|Input tensor 4D +|Input|in_t*|input|[N,IH,IW,C]|Input tensor 4D |Attribute|int*|kernel|[2]|[kernel_y, kernel_x] |Attribute|int*|stride|[2]|[stride_y, stride_x] |Attribute|int*|pad|[4]|[pad_top, pad_bottom, pad_left, pad_right] @@ -448,13 +452,13 @@ for_each(0 <= n < N, 0 <= oy < H, 0 <= ox < W, 0 <= c < C ) { in_t acc = minimum_value<in_t>; iy = oy * stride_y - pad_top; ix = ox * stride_x - pad_left; - for_each( 0<=ky<kernel_y, 0<=kx<kernel_x ) { + for_each( 0 <= ky < kernel_y, 0 <= kx < kernel_x ) { y = iy + ky; x = ix + kx; - in_t value = tensor_read<in_t>(input, [N,IH,IW,IC], [n,y,x,c], pad); + in_t value = tensor_read<in_t>(input, [N,IH,IW,C], [n,y,x,c], pad); acc = apply_max(acc, value); } - tensor_write<in_t>(output, [N,H,W,OC], [n,oy,ox,oc], acc); + tensor_write<in_t>(output, [N,H,W,C], [n,oy,ox,c], acc); } ---- |