MrBesen/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

lavfi/dnn: refine dnn interface to add DNNExecBaseParams

Browse Source 
Different function type of model requires different parameters, for
example, object detection detects lots of objects (cat/dog/...) in
the frame, and classifcation needs to know which object (cat or dog)
it is going to classify.

The current interface needs to add a new function with more parameters
to support new requirement, with this change, we can just add a new
struct (for example DNNExecClassifyParams) based on DNNExecBaseParams,
and so we can continue to use the current interface execute_model just
with params changed.
This commit is contained in:
Guo, Yejun 2021-04-01 10:06:06 +08:00
parent 7eb9accc37
commit a3b74651a0
11 changed files with 139 additions and 73 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
libavfilter/dnn/Makefile
View File

@ -2,6 +2,7 @@ OBJS-$(CONFIG_DNN) += dnn/dnn_interface.o
OBJS-$(CONFIG_DNN) += dnn/dnn_io_proc.o
OBJS-$(CONFIG_DNN) += dnn/queue.o
OBJS-$(CONFIG_DNN) += dnn/safe_queue.o
OBJS-$(CONFIG_DNN) += dnn/dnn_backend_common.o
OBJS-$(CONFIG_DNN) += dnn/dnn_backend_native.o
OBJS-$(CONFIG_DNN) += dnn/dnn_backend_native_layers.o
OBJS-$(CONFIG_DNN) += dnn/dnn_backend_native_layer_avgpool.o

51
libavfilter/dnn/dnn_backend_common.c Normal file
View File

@ -0,0 +1,51 @@
/*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* DNN common functions different backends.
*/
#include "dnn_backend_common.h"
int ff_check_exec_params(void *ctx, DNNBackendType backend, DNNFunctionType func_type, DNNExecBaseParams *exec_params)
{
if (!exec_params) {
av_log(ctx, AV_LOG_ERROR, "exec_params is null when execute model.\n");
return AVERROR(EINVAL);
}
if (!exec_params->in_frame) {
av_log(ctx, AV_LOG_ERROR, "in frame is NULL when execute model.\n");
return AVERROR(EINVAL);
}
if (!exec_params->out_frame) {
av_log(ctx, AV_LOG_ERROR, "out frame is NULL when execute model.\n");
return AVERROR(EINVAL);
}
if (exec_params->nb_output != 1 && backend != DNN_TF) {
// currently, the filter does not need multiple outputs,
// so we just pending the support until we really need it.
avpriv_report_missing_feature(ctx, "multiple outputs");
return AVERROR(EINVAL);
}
return 0;
}

31
libavfilter/dnn/dnn_backend_common.h Normal file
View File

@ -0,0 +1,31 @@
/*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* DNN common functions different backends.
*/
#ifndef AVFILTER_DNN_DNN_BACKEND_COMMON_H
#define AVFILTER_DNN_DNN_BACKEND_COMMON_H
#include "../dnn_interface.h"
int ff_check_exec_params(void *ctx, DNNBackendType backend, DNNFunctionType func_type, DNNExecBaseParams *exec_params);
#endif

15
libavfilter/dnn/dnn_backend_native.c
View File

@ -28,6 +28,7 @@
#include "dnn_backend_native_layer_conv2d.h"
#include "dnn_backend_native_layers.h"
#include "dnn_io_proc.h"
#include "dnn_backend_common.h"
#define OFFSET(x) offsetof(NativeContext, x)
#define FLAGS AV_OPT_FLAG_FILTERING_PARAM
@ -372,23 +373,17 @@ static DNNReturnType execute_model_native(const DNNModel *model, const char *inp
return DNN_SUCCESS;
}
DNNReturnType ff_dnn_execute_model_native(const DNNModel *model, const char *input_name, AVFrame *in_frame,
const char **output_names, uint32_t nb_output, AVFrame *out_frame)
DNNReturnType ff_dnn_execute_model_native(const DNNModel *model, DNNExecBaseParams *exec_params)
{
NativeModel *native_model = model->model;
NativeContext *ctx = &native_model->ctx;
if (!in_frame) {
av_log(ctx, AV_LOG_ERROR, "in frame is NULL when execute model.\n");
if (ff_check_exec_params(ctx, DNN_NATIVE, model->func_type, exec_params) != 0) {
return DNN_ERROR;
}
if (!out_frame) {
av_log(ctx, AV_LOG_ERROR, "out frame is NULL when execute model.\n");
return DNN_ERROR;
}
return execute_model_native(model, input_name, in_frame, output_names, nb_output, out_frame, 1);
return execute_model_native(model, exec_params->input_name, exec_params->in_frame,
exec_params->output_names, exec_params->nb_output, exec_params->out_frame, 1);
}
int32_t ff_calculate_operand_dims_count(const DnnOperand *oprd)

3
libavfilter/dnn/dnn_backend_native.h
View File

@ -130,8 +130,7 @@ typedef struct NativeModel{
DNNModel *ff_dnn_load_model_native(const char *model_filename, DNNFunctionType func_type, const char *options, AVFilterContext *filter_ctx);
DNNReturnType ff_dnn_execute_model_native(const DNNModel *model, const char *input_name, AVFrame *in_frame,
const char **output_names, uint32_t nb_output, AVFrame *out_frame);
DNNReturnType ff_dnn_execute_model_native(const DNNModel *model, DNNExecBaseParams *exec_params);
void ff_dnn_free_model_native(DNNModel **model);

50
libavfilter/dnn/dnn_backend_openvino.c
View File

@ -33,6 +33,7 @@
#include "queue.h"
#include "safe_queue.h"
#include <c_api/ie_c_api.h>
#include "dnn_backend_common.h"
typedef struct OVOptions{
char *device_type;
@ -678,28 +679,14 @@ err:
return NULL;
}
DNNReturnType ff_dnn_execute_model_ov(const DNNModel *model, const char *input_name, AVFrame *in_frame,
const char **output_names, uint32_t nb_output, AVFrame *out_frame)
DNNReturnType ff_dnn_execute_model_ov(const DNNModel *model, DNNExecBaseParams *exec_params)
{
OVModel *ov_model = model->model;
OVContext *ctx = &ov_model->ctx;
TaskItem task;
RequestItem *request;
if (!in_frame) {
av_log(ctx, AV_LOG_ERROR, "in frame is NULL when execute model.\n");
return DNN_ERROR;
}
if (!out_frame && model->func_type == DFT_PROCESS_FRAME) {
av_log(ctx, AV_LOG_ERROR, "out frame is NULL when execute model.\n");
return DNN_ERROR;
}
if (nb_output != 1) {
// currently, the filter does not need multiple outputs,
// so we just pending the support until we really need it.
avpriv_report_missing_feature(ctx, "multiple outputs");
if (ff_check_exec_params(ctx, DNN_OV, model->func_type, exec_params) != 0) {
return DNN_ERROR;
}
@ -709,7 +696,7 @@ DNNReturnType ff_dnn_execute_model_ov(const DNNModel *model, const char *input_n
}
if (!ov_model->exe_network) {
if (init_model_ov(ov_model, input_name, output_names[0]) != DNN_SUCCESS) {
if (init_model_ov(ov_model, exec_params->input_name, exec_params->output_names[0]) != DNN_SUCCESS) {
av_log(ctx, AV_LOG_ERROR, "Failed init OpenVINO exectuable network or inference request\n");
return DNN_ERROR;
}
@ -717,10 +704,10 @@ DNNReturnType ff_dnn_execute_model_ov(const DNNModel *model, const char *input_n
task.do_ioproc = 1;
task.async = 0;
task.input_name = input_name;
task.in_frame = in_frame;
task.output_name = output_names[0];
task.out_frame = out_frame;
task.input_name = exec_params->input_name;
task.in_frame = exec_params->in_frame;
task.output_name = exec_params->output_names[0];
task.out_frame = exec_params->out_frame ? exec_params->out_frame : exec_params->in_frame;
task.ov_model = ov_model;
if (extract_inference_from_task(ov_model->model->func_type, &task, ov_model->inference_queue) != DNN_SUCCESS) {
@ -737,26 +724,19 @@ DNNReturnType ff_dnn_execute_model_ov(const DNNModel *model, const char *input_n
return execute_model_ov(request, ov_model->inference_queue);
}
DNNReturnType ff_dnn_execute_model_async_ov(const DNNModel *model, const char *input_name, AVFrame *in_frame,
const char **output_names, uint32_t nb_output, AVFrame *out_frame)
DNNReturnType ff_dnn_execute_model_async_ov(const DNNModel *model, DNNExecBaseParams *exec_params)
{
OVModel *ov_model = model->model;
OVContext *ctx = &ov_model->ctx;
RequestItem *request;
TaskItem *task;
if (!in_frame) {
av_log(ctx, AV_LOG_ERROR, "in frame is NULL when async execute model.\n");
return DNN_ERROR;
}
if (!out_frame && model->func_type == DFT_PROCESS_FRAME) {
av_log(ctx, AV_LOG_ERROR, "out frame is NULL when async execute model.\n");
if (ff_check_exec_params(ctx, DNN_OV, model->func_type, exec_params) != 0) {
return DNN_ERROR;
}
if (!ov_model->exe_network) {
if (init_model_ov(ov_model, input_name, output_names[0]) != DNN_SUCCESS) {
if (init_model_ov(ov_model, exec_params->input_name, exec_params->output_names[0]) != DNN_SUCCESS) {
av_log(ctx, AV_LOG_ERROR, "Failed init OpenVINO exectuable network or inference request\n");
return DNN_ERROR;
}
@ -770,10 +750,10 @@ DNNReturnType ff_dnn_execute_model_async_ov(const DNNModel *model, const char *i
task->do_ioproc = 1;
task->async = 1;
task->input_name = input_name;
task->in_frame = in_frame;
task->output_name = output_names[0];
task->out_frame = out_frame;
task->input_name = exec_params->input_name;
task->in_frame = exec_params->in_frame;
task->output_name = exec_params->output_names[0];
task->out_frame = exec_params->out_frame ? exec_params->out_frame : exec_params->in_frame;
task->ov_model = ov_model;
if (ff_queue_push_back(ov_model->task_queue, task) < 0) {
av_freep(&task);

6
libavfilter/dnn/dnn_backend_openvino.h
View File

@ -31,10 +31,8 @@
DNNModel *ff_dnn_load_model_ov(const char *model_filename, DNNFunctionType func_type, const char *options, AVFilterContext *filter_ctx);
DNNReturnType ff_dnn_execute_model_ov(const DNNModel *model, const char *input_name, AVFrame *in_frame,
const char **output_names, uint32_t nb_output, AVFrame *out_frame);
DNNReturnType ff_dnn_execute_model_async_ov(const DNNModel *model, const char *input_name, AVFrame *in_frame,
const char **output_names, uint32_t nb_output, AVFrame *out_frame);
DNNReturnType ff_dnn_execute_model_ov(const DNNModel *model, DNNExecBaseParams *exec_params);
DNNReturnType ff_dnn_execute_model_async_ov(const DNNModel *model, DNNExecBaseParams *exec_params);
DNNAsyncStatusType ff_dnn_get_async_result_ov(const DNNModel *model, AVFrame **in, AVFrame **out);
DNNReturnType ff_dnn_flush_ov(const DNNModel *model);

18
libavfilter/dnn/dnn_backend_tf.c
View File

@ -34,7 +34,7 @@
#include "dnn_backend_native_layer_pad.h"
#include "dnn_backend_native_layer_maximum.h"
#include "dnn_io_proc.h"
#include "dnn_backend_common.h"
#include <tensorflow/c/c_api.h>
typedef struct TFOptions{
@ -814,23 +814,17 @@ static DNNReturnType execute_model_tf(const DNNModel *model, const char *input_n
return DNN_SUCCESS;
}
DNNReturnType ff_dnn_execute_model_tf(const DNNModel *model, const char *input_name, AVFrame *in_frame,
const char **output_names, uint32_t nb_output, AVFrame *out_frame)
DNNReturnType ff_dnn_execute_model_tf(const DNNModel *model, DNNExecBaseParams *exec_params)
{
TFModel *tf_model = model->model;
TFContext *ctx = &tf_model->ctx;
if (!in_frame) {
av_log(ctx, AV_LOG_ERROR, "in frame is NULL when execute model.\n");
return DNN_ERROR;
if (ff_check_exec_params(ctx, DNN_TF, model->func_type, exec_params) != 0) {
return DNN_ERROR;
}
if (!out_frame) {
av_log(ctx, AV_LOG_ERROR, "out frame is NULL when execute model.\n");
return DNN_ERROR;
}
return execute_model_tf(model, input_name, in_frame, output_names, nb_output, out_frame, 1);
return execute_model_tf(model, exec_params->input_name, exec_params->in_frame,
exec_params->output_names, exec_params->nb_output, exec_params->out_frame, 1);
}
void ff_dnn_free_model_tf(DNNModel **model)

3
libavfilter/dnn/dnn_backend_tf.h
View File

@ -31,8 +31,7 @@
DNNModel *ff_dnn_load_model_tf(const char *model_filename, DNNFunctionType func_type, const char *options, AVFilterContext *filter_ctx);
DNNReturnType ff_dnn_execute_model_tf(const DNNModel *model, const char *input_name, AVFrame *in_frame,
const char **output_names, uint32_t nb_output, AVFrame *out_frame);
DNNReturnType ff_dnn_execute_model_tf(const DNNModel *model, DNNExecBaseParams *exec_params);
void ff_dnn_free_model_tf(DNNModel **model);

20
libavfilter/dnn_filter_common.c
View File

@ -90,14 +90,26 @@ DNNReturnType ff_dnn_get_output(DnnContext *ctx, int input_width, int input_heig
DNNReturnType ff_dnn_execute_model(DnnContext *ctx, AVFrame *in_frame, AVFrame *out_frame)
{
return (ctx->dnn_module->execute_model)(ctx->model, ctx->model_inputname, in_frame,
(const char **)&ctx->model_outputname, 1, out_frame);
DNNExecBaseParams exec_params = {
.input_name = ctx->model_inputname,
.output_names = (const char **)&ctx->model_outputname,
.nb_output = 1,
.in_frame = in_frame,
.out_frame = out_frame,
};
return (ctx->dnn_module->execute_model)(ctx->model, &exec_params);
}
DNNReturnType ff_dnn_execute_model_async(DnnContext *ctx, AVFrame *in_frame, AVFrame *out_frame)
{
return (ctx->dnn_module->execute_model_async)(ctx->model, ctx->model_inputname, in_frame,
(const char **)&ctx->model_outputname, 1, out_frame);
DNNExecBaseParams exec_params = {
.input_name = ctx->model_inputname,
.output_names = (const char **)&ctx->model_outputname,
.nb_output = 1,
.in_frame = in_frame,
.out_frame = out_frame,
};
return (ctx->dnn_module->execute_model_async)(ctx->model, &exec_params);
}
DNNAsyncStatusType ff_dnn_get_async_result(DnnContext *ctx, AVFrame **in_frame, AVFrame **out_frame)

14
libavfilter/dnn_interface.h
View File

@ -63,6 +63,14 @@ typedef struct DNNData{
DNNColorOrder order;
} DNNData;
typedef struct DNNExecBaseParams {
const char *input_name;
const char **output_names;
uint32_t nb_output;
AVFrame *in_frame;
AVFrame *out_frame;
} DNNExecBaseParams;
typedef int (*FramePrePostProc)(AVFrame *frame, DNNData *model, AVFilterContext *filter_ctx);
typedef int (*DetectPostProc)(AVFrame *frame, DNNData *output, uint32_t nb, AVFilterContext *filter_ctx);
@ -96,11 +104,9 @@ typedef struct DNNModule{
// Loads model and parameters from given file. Returns NULL if it is not possible.
DNNModel *(*load_model)(const char *model_filename, DNNFunctionType func_type, const char *options, AVFilterContext *filter_ctx);
// Executes model with specified input and output. Returns DNN_ERROR otherwise.
DNNReturnType (*execute_model)(const DNNModel *model, const char *input_name, AVFrame *in_frame,
const char **output_names, uint32_t nb_output, AVFrame *out_frame);
DNNReturnType (*execute_model)(const DNNModel *model, DNNExecBaseParams *exec_params);
// Executes model with specified input and output asynchronously. Returns DNN_ERROR otherwise.
DNNReturnType (*execute_model_async)(const DNNModel *model, const char *input_name, AVFrame *in_frame,
const char **output_names, uint32_t nb_output, AVFrame *out_frame);
DNNReturnType (*execute_model_async)(const DNNModel *model, DNNExecBaseParams *exec_params);
// Retrieve inference result.
DNNAsyncStatusType (*get_async_result)(const DNNModel *model, AVFrame **in, AVFrame **out);
// Flush all the pending tasks.