2 * Copyright (c) 2018 Sergey Lavrushkin
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * DNN inference functions interface for native backend.
27 #ifndef AVFILTER_DNN_DNN_BACKEND_NATIVE_H
28 #define AVFILTER_DNN_DNN_BACKEND_NATIVE_H
30 #include "../dnn_interface.h"
31 #include "libavformat/avio.h"
33 typedef enum {INPUT, CONV, DEPTH_TO_SPACE, MIRROR_PAD} DNNLayerType;
35 typedef enum {DOT_INPUT = 1, DOT_OUTPUT = 2, DOT_INTERMEDIATE = DOT_INPUT | DOT_INPUT} DNNOperandType;
40 * a layer can have multiple inputs and one output.
41 * 4 is just a big enough number for input operands (increase it if necessary),
42 * do not use 'int32_t *input_operand_indexes', so we don't worry about mem leaks.
44 int32_t input_operand_indexes[4];
45 int32_t output_operand_index;
49 typedef struct DnnOperand{
51 * there are two memory layouts, NHWC or NCHW, so we use dims,
57 * input/output/intermediate operand of the network
62 * support different kinds of data type such as float, half float, int8 etc,
63 * first support float now.
65 DNNDataType data_type;
68 * NHWC if 1, otherwise NCHW.
69 * let's first support NHWC only, this flag is for extensive usage.
74 * to avoid possible memory leak, do not use char *name
79 * data pointer with data length in bytes.
80 * usedNumbersLeft is only valid for intermediate operand,
81 * it means how many layers still depend on this operand,
82 * todo: the memory can be reused when usedNumbersLeft is zero.
86 int32_t usedNumbersLeft;
89 typedef struct InputParams{
90 int height, width, channels;
93 // Represents simple feed-forward convolutional network.
94 typedef struct ConvolutionalNetwork{
99 } ConvolutionalNetwork;
101 DNNModel *ff_dnn_load_model_native(const char *model_filename);
103 DNNReturnType ff_dnn_execute_model_native(const DNNModel *model, DNNData *outputs, uint32_t nb_output);
105 void ff_dnn_free_model_native(DNNModel **model);
107 int32_t calculate_operand_data_length(DnnOperand *operand);