FFmpeg
dnn_filter_common.c
Go to the documentation of this file.
1 /*
2  * This file is part of FFmpeg.
3  *
4  * FFmpeg is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU Lesser General Public
6  * License as published by the Free Software Foundation; either
7  * version 2.1 of the License, or (at your option) any later version.
8  *
9  * FFmpeg is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12  * Lesser General Public License for more details.
13  *
14  * You should have received a copy of the GNU Lesser General Public
15  * License along with FFmpeg; if not, write to the Free Software
16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17  */
18 
19 #include "dnn_filter_common.h"
20 #include "libavutil/avstring.h"
21 
22 #define MAX_SUPPORTED_OUTPUTS_NB 4
23 
24 static char **separate_output_names(const char *expr, const char *val_sep, int *separated_nb)
25 {
26  char *val, **parsed_vals = NULL;
27  int val_num = 0;
28  if (!expr || !val_sep || !separated_nb) {
29  return NULL;
30  }
31 
32  parsed_vals = av_calloc(MAX_SUPPORTED_OUTPUTS_NB, sizeof(*parsed_vals));
33  if (!parsed_vals) {
34  return NULL;
35  }
36 
37  do {
38  val = av_get_token(&expr, val_sep);
39  if(val) {
40  parsed_vals[val_num] = val;
41  val_num++;
42  }
43  if (*expr) {
44  expr++;
45  }
46  } while(*expr);
47 
48  parsed_vals[val_num] = NULL;
49  *separated_nb = val_num;
50 
51  return parsed_vals;
52 }
53 
55 {
56  if (!ctx->model_filename) {
57  av_log(filter_ctx, AV_LOG_ERROR, "model file for network is not specified\n");
58  return AVERROR(EINVAL);
59  }
60  if (!ctx->model_inputname) {
61  av_log(filter_ctx, AV_LOG_ERROR, "input name of the model network is not specified\n");
62  return AVERROR(EINVAL);
63  }
64 
65  ctx->model_outputnames = separate_output_names(ctx->model_outputnames_string, "&", &ctx->nb_outputs);
66  if (!ctx->model_outputnames) {
67  av_log(filter_ctx, AV_LOG_ERROR, "could not parse model output names\n");
68  return AVERROR(EINVAL);
69  }
70 
71  ctx->dnn_module = ff_get_dnn_module(ctx->backend_type, filter_ctx);
72  if (!ctx->dnn_module) {
73  av_log(filter_ctx, AV_LOG_ERROR, "could not create DNN module for requested backend\n");
74  return AVERROR(ENOMEM);
75  }
76  if (!ctx->dnn_module->load_model) {
77  av_log(filter_ctx, AV_LOG_ERROR, "load_model for network is not specified\n");
78  return AVERROR(EINVAL);
79  }
80 
81  ctx->model = (ctx->dnn_module->load_model)(ctx->model_filename, func_type, ctx->backend_options, filter_ctx);
82  if (!ctx->model) {
83  av_log(filter_ctx, AV_LOG_ERROR, "could not load DNN model\n");
84  return AVERROR(EINVAL);
85  }
86 
87  return 0;
88 }
89 
91 {
92  ctx->model->frame_pre_proc = pre_proc;
93  ctx->model->frame_post_proc = post_proc;
94  return 0;
95 }
96 
98 {
99  ctx->model->detect_post_proc = post_proc;
100  return 0;
101 }
102 
104 {
105  ctx->model->classify_post_proc = post_proc;
106  return 0;
107 }
108 
110 {
111  return ctx->model->get_input(ctx->model->model, input, ctx->model_inputname);
112 }
113 
114 int ff_dnn_get_output(DnnContext *ctx, int input_width, int input_height, int *output_width, int *output_height)
115 {
116  return ctx->model->get_output(ctx->model->model, ctx->model_inputname, input_width, input_height,
117  (const char *)ctx->model_outputnames[0], output_width, output_height);
118 }
119 
120 int ff_dnn_execute_model(DnnContext *ctx, AVFrame *in_frame, AVFrame *out_frame)
121 {
122  DNNExecBaseParams exec_params = {
123  .input_name = ctx->model_inputname,
124  .output_names = (const char **)ctx->model_outputnames,
125  .nb_output = ctx->nb_outputs,
126  .in_frame = in_frame,
127  .out_frame = out_frame,
128  };
129  return (ctx->dnn_module->execute_model)(ctx->model, &exec_params);
130 }
131 
132 int ff_dnn_execute_model_classification(DnnContext *ctx, AVFrame *in_frame, AVFrame *out_frame, const char *target)
133 {
134  DNNExecClassificationParams class_params = {
135  {
136  .input_name = ctx->model_inputname,
137  .output_names = (const char **)ctx->model_outputnames,
138  .nb_output = ctx->nb_outputs,
139  .in_frame = in_frame,
140  .out_frame = out_frame,
141  },
142  .target = target,
143  };
144  return (ctx->dnn_module->execute_model)(ctx->model, &class_params.base);
145 }
146 
148 {
149  return (ctx->dnn_module->get_result)(ctx->model, in_frame, out_frame);
150 }
151 
153 {
154  return (ctx->dnn_module->flush)(ctx->model);
155 }
156 
158 {
159  if (ctx->dnn_module) {
160  (ctx->dnn_module->free_model)(&ctx->model);
161  }
162  if (ctx->model_outputnames) {
163  for (int i = 0; i < ctx->nb_outputs; i++)
164  av_free(ctx->model_outputnames[i]);
165 
166  av_freep(&ctx->model_outputnames);
167  }
168 }
MAX_SUPPORTED_OUTPUTS_NB
#define MAX_SUPPORTED_OUTPUTS_NB
Definition: dnn_filter_common.c:22
AVERROR
Filter the word “frame” indicates either a video frame or a group of audio as stored in an AVFrame structure Format for each input and each output the list of supported formats For video that means pixel format For audio that means channel sample they are references to shared objects When the negotiation mechanism computes the intersection of the formats supported at each end of a all references to both lists are replaced with a reference to the intersection And when a single format is eventually chosen for a link amongst the remaining all references to the list are updated That means that if a filter requires that its input and output have the same format amongst a supported all it has to do is use a reference to the same list of formats query_formats can leave some formats unset and return AVERROR(EAGAIN) to cause the negotiation mechanism toagain later. That can be used by filters with complex requirements to use the format negotiated on one link to set the formats supported on another. Frame references ownership and permissions
DNNFunctionType
DNNFunctionType
Definition: dnn_interface.h:52
AVFrame
This structure describes decoded (raw) audio or video data.
Definition: frame.h:340
DetectPostProc
int(* DetectPostProc)(AVFrame *frame, DNNData *output, uint32_t nb, AVFilterContext *filter_ctx)
Definition: dnn_interface.h:90
separate_output_names
static char ** separate_output_names(const char *expr, const char *val_sep, int *separated_nb)
Definition: dnn_filter_common.c:24
DNNExecBaseParams::input_name
const char * input_name
Definition: dnn_interface.h:77
dnn_filter_common.h
DnnContext
Definition: dnn_filter_common.h:29
filter_ctx
static FilteringContext * filter_ctx
Definition: transcode.c:51
val
static double val(void *priv, double ch)
Definition: aeval.c:78
AV_LOG_ERROR
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:180
DNNExecClassificationParams
Definition: dnn_interface.h:84
ff_dnn_set_detect_post_proc
int ff_dnn_set_detect_post_proc(DnnContext *ctx, DetectPostProc post_proc)
Definition: dnn_filter_common.c:97
DNNData
Definition: dnn_interface.h:65
ff_dnn_get_result
DNNAsyncStatusType ff_dnn_get_result(DnnContext *ctx, AVFrame **in_frame, AVFrame **out_frame)
Definition: dnn_filter_common.c:147
ctx
AVFormatContext * ctx
Definition: movenc.c:48
ff_dnn_get_input
int ff_dnn_get_input(DnnContext *ctx, DNNData *input)
Definition: dnn_filter_common.c:109
DNNExecClassificationParams::base
DNNExecBaseParams base
Definition: dnn_interface.h:85
NULL
#define NULL
Definition: coverity.c:32
ff_dnn_execute_model_classification
int ff_dnn_execute_model_classification(DnnContext *ctx, AVFrame *in_frame, AVFrame *out_frame, const char *target)
Definition: dnn_filter_common.c:132
ff_dnn_flush
int ff_dnn_flush(DnnContext *ctx)
Definition: dnn_filter_common.c:152
ClassifyPostProc
int(* ClassifyPostProc)(AVFrame *frame, DNNData *output, uint32_t bbox_index, AVFilterContext *filter_ctx)
Definition: dnn_interface.h:91
ff_get_dnn_module
const DNNModule * ff_get_dnn_module(DNNBackendType backend_type, void *log_ctx)
Definition: dnn_interface.c:32
input
and forward the test the status of outputs and forward it to the corresponding return FFERROR_NOT_READY If the filters stores internally one or a few frame for some input
Definition: filter_design.txt:172
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:255
ff_dnn_set_frame_proc
int ff_dnn_set_frame_proc(DnnContext *ctx, FramePrePostProc pre_proc, FramePrePostProc post_proc)
Definition: dnn_filter_common.c:90
av_calloc
void * av_calloc(size_t nmemb, size_t size)
Definition: mem.c:262
ff_dnn_get_output
int ff_dnn_get_output(DnnContext *ctx, int input_width, int input_height, int *output_width, int *output_height)
Definition: dnn_filter_common.c:114
av_get_token
char * av_get_token(const char **buf, const char *term)
Unescape the given string until a non escaped terminating char, and return the token corresponding to...
Definition: avstring.c:143
FramePrePostProc
int(* FramePrePostProc)(AVFrame *frame, DNNData *model, AVFilterContext *filter_ctx)
Definition: dnn_interface.h:89
AVFilterContext
An instance of a filter.
Definition: avfilter.h:397
av_free
#define av_free(p)
Definition: tableprint_vlc.h:33
ff_dnn_init
int ff_dnn_init(DnnContext *ctx, DNNFunctionType func_type, AVFilterContext *filter_ctx)
Definition: dnn_filter_common.c:54
av_freep
#define av_freep(p)
Definition: tableprint_vlc.h:34
DNNExecBaseParams
Definition: dnn_interface.h:76
av_log
#define av_log(a,...)
Definition: tableprint_vlc.h:27
ff_dnn_uninit
void ff_dnn_uninit(DnnContext *ctx)
Definition: dnn_filter_common.c:157
ff_dnn_execute_model
int ff_dnn_execute_model(DnnContext *ctx, AVFrame *in_frame, AVFrame *out_frame)
Definition: dnn_filter_common.c:120
avstring.h
DNNAsyncStatusType
DNNAsyncStatusType
Definition: dnn_interface.h:45
ff_dnn_set_classify_post_proc
int ff_dnn_set_classify_post_proc(DnnContext *ctx, ClassifyPostProc post_proc)
Definition: dnn_filter_common.c:103