dnn_backend_native_layer_avgpool.c File Reference

#include "libavutil/avassert.h"
#include "dnn_backend_native_layer_avgpool.h"

Go to the source code of this file.

Functions
int	ff_dnn_load_layer_avg_pool (Layer *layer, AVIOContext *model_file_context, int file_size, int operands_num)
	Load Average Pooling Layer. More...
int	ff_dnn_execute_layer_avg_pool (DnnOperand *operands, const int32_t *input_operand_indexes, int32_t output_operand_index, const void *parameters, NativeContext *ctx)
	Execute the Average Pooling Layer. More...

Detailed Description

DNN native backend implementation.

Definition in file dnn_backend_native_layer_avgpool.c.

Function Documentation

ff_dnn_load_layer_avg_pool()

int ff_dnn_load_layer_avg_pool	(	Layer *	layer,
		AVIOContext *	model_file_context,
		int	file_size,
		int	operands_num
	)

Load Average Pooling Layer.

It assigns the Average Pooling layer with AvgPoolParams after parsing from the model file context.

Parameters

layer	pointer to the DNN layer instance
model_file_context	pointer to model file context
file_size	model file size to check if data is read correctly from the model file
operands_num	operand count of the whole model to check if data is read correctly from the model file

Returns

number of bytes read from the model file

Return values

0	if out of memory or an error occurs

Definition at line 29 of file dnn_backend_native_layer_avgpool.c.

ff_dnn_execute_layer_avg_pool()

int ff_dnn_execute_layer_avg_pool	(	DnnOperand *	operands,
		const int32_t *	input_operand_indexes,
		int32_t	output_operand_index,
		const void *	parameters,
		NativeContext *	ctx
	)

Execute the Average Pooling Layer.

Padding in channel dimensions is currently not supported.

Parameters

operands	all operands for the model
input_operand_indexes	input operand indexes for this layer
output_operand_index	output operand index for this layer
parameters	average pooling parameters
ctx	pointer to Native model context for logging

Return values

0	if the execution succeeds
DNN_ERROR	if the execution fails

When padding_method = SAME, the tensorflow will only padding the hald number of 0 pixels except the remainders. Eg: assuming the input height = 1080, the strides = 11, so the remainders = 1080 % 11 = 2 and if ksize = 5: it will fill (5 - 2) >> 1 = 1 line before the first line of input image, and 5 - 2 - 1 = 2 lines after the last line of input image. and if ksize = 7: it will fill (7 - 2) >> 1 = 2 lines before the first line of input image, and 7 - 2 - 2 = 3 lines after the last line of input image.

Definition at line 58 of file dnn_backend_native_layer_avgpool.c.

Referenced by test_with_same(), and test_with_valid().