1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
|
/*
* Copyright (c) 2019 Guo Yejun
*
* 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
* implementing a generic image processing filter using deep learning networks.
*/
#include "libavformat/avio.h"
#include "libavutil/opt.h"
#include "libavutil/pixdesc.h"
#include "libavutil/avassert.h"
#include "avfilter.h"
#include "dnn_interface.h"
#include "formats.h"
#include "internal.h"
typedef struct DnnProcessingContext {
const AVClass *class;
char *model_filename;
DNNBackendType backend_type;
char *model_inputname;
char *model_outputname;
DNNModule *dnn_module;
DNNModel *model;
// input & output of the model at execution time
DNNData input;
DNNData output;
} DnnProcessingContext;
#define OFFSET(x) offsetof(DnnProcessingContext, x)
#define FLAGS AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM
static const AVOption dnn_processing_options[] = {
{ "dnn_backend", "DNN backend", OFFSET(backend_type), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS, "backend" },
{ "native", "native backend flag", 0, AV_OPT_TYPE_CONST, { .i64 = 0 }, 0, 0, FLAGS, "backend" },
#if (CONFIG_LIBTENSORFLOW == 1)
{ "tensorflow", "tensorflow backend flag", 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, 0, 0, FLAGS, "backend" },
#endif
{ "model", "path to model file", OFFSET(model_filename), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, FLAGS },
{ "input", "input name of the model", OFFSET(model_inputname), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, FLAGS },
{ "output", "output name of the model", OFFSET(model_outputname), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, FLAGS },
{ NULL }
};
AVFILTER_DEFINE_CLASS(dnn_processing);
static av_cold int init(AVFilterContext *context)
{
DnnProcessingContext *ctx = context->priv;
if (!ctx->model_filename) {
av_log(ctx, AV_LOG_ERROR, "model file for network is not specified\n");
return AVERROR(EINVAL);
}
if (!ctx->model_inputname) {
av_log(ctx, AV_LOG_ERROR, "input name of the model network is not specified\n");
return AVERROR(EINVAL);
}
if (!ctx->model_outputname) {
av_log(ctx, AV_LOG_ERROR, "output name of the model network is not specified\n");
return AVERROR(EINVAL);
}
ctx->dnn_module = ff_get_dnn_module(ctx->backend_type);
if (!ctx->dnn_module) {
av_log(ctx, AV_LOG_ERROR, "could not create DNN module for requested backend\n");
return AVERROR(ENOMEM);
}
if (!ctx->dnn_module->load_model) {
av_log(ctx, AV_LOG_ERROR, "load_model for network is not specified\n");
return AVERROR(EINVAL);
}
ctx->model = (ctx->dnn_module->load_model)(ctx->model_filename);
if (!ctx->model) {
av_log(ctx, AV_LOG_ERROR, "could not load DNN model\n");
return AVERROR(EINVAL);
}
return 0;
}
static int query_formats(AVFilterContext *context)
{
static const enum AVPixelFormat pix_fmts[] = {
AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24,
AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAYF32,
AV_PIX_FMT_NONE
};
AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
return ff_set_common_formats(context, fmts_list);
}
#define LOG_FORMAT_CHANNEL_MISMATCH() \
av_log(ctx, AV_LOG_ERROR, \
"the frame's format %s does not match " \
"the model input channel %d\n", \
av_get_pix_fmt_name(fmt), \
model_input->channels);
static int check_modelinput_inlink(const DNNData *model_input, const AVFilterLink *inlink)
{
AVFilterContext *ctx = inlink->dst;
enum AVPixelFormat fmt = inlink->format;
// the design is to add explicit scale filter before this filter
if (model_input->height != -1 && model_input->height != inlink->h) {
av_log(ctx, AV_LOG_ERROR, "the model requires frame height %d but got %d\n",
model_input->height, inlink->h);
return AVERROR(EIO);
}
if (model_input->width != -1 && model_input->width != inlink->w) {
av_log(ctx, AV_LOG_ERROR, "the model requires frame width %d but got %d\n",
model_input->width, inlink->w);
return AVERROR(EIO);
}
switch (fmt) {
case AV_PIX_FMT_RGB24:
case AV_PIX_FMT_BGR24:
if (model_input->channels != 3) {
LOG_FORMAT_CHANNEL_MISMATCH();
return AVERROR(EIO);
}
if (model_input->dt != DNN_FLOAT && model_input->dt != DNN_UINT8) {
av_log(ctx, AV_LOG_ERROR, "only support dnn models with input data type as float32 and uint8.\n");
return AVERROR(EIO);
}
return 0;
case AV_PIX_FMT_GRAY8:
if (model_input->channels != 1) {
LOG_FORMAT_CHANNEL_MISMATCH();
return AVERROR(EIO);
}
if (model_input->dt != DNN_UINT8) {
av_log(ctx, AV_LOG_ERROR, "only support dnn models with input data type uint8.\n");
return AVERROR(EIO);
}
return 0;
case AV_PIX_FMT_GRAYF32:
if (model_input->channels != 1) {
LOG_FORMAT_CHANNEL_MISMATCH();
return AVERROR(EIO);
}
if (model_input->dt != DNN_FLOAT) {
av_log(ctx, AV_LOG_ERROR, "only support dnn models with input data type float32.\n");
return AVERROR(EIO);
}
return 0;
default:
av_log(ctx, AV_LOG_ERROR, "%s not supported.\n", av_get_pix_fmt_name(fmt));
return AVERROR(EIO);
}
return 0;
}
static int config_input(AVFilterLink *inlink)
{
AVFilterContext *context = inlink->dst;
DnnProcessingContext *ctx = context->priv;
DNNReturnType result;
DNNData model_input;
int check;
result = ctx->model->get_input(ctx->model->model, &model_input, ctx->model_inputname);
if (result != DNN_SUCCESS) {
av_log(ctx, AV_LOG_ERROR, "could not get input from the model\n");
return AVERROR(EIO);
}
check = check_modelinput_inlink(&model_input, inlink);
if (check != 0) {
return check;
}
ctx->input.width = inlink->w;
ctx->input.height = inlink->h;
ctx->input.channels = model_input.channels;
ctx->input.dt = model_input.dt;
result = (ctx->model->set_input_output)(ctx->model->model,
&ctx->input, ctx->model_inputname,
(const char **)&ctx->model_outputname, 1);
if (result != DNN_SUCCESS) {
av_log(ctx, AV_LOG_ERROR, "could not set input and output for the model\n");
return AVERROR(EIO);
}
return 0;
}
static int config_output(AVFilterLink *outlink)
{
AVFilterContext *context = outlink->src;
DnnProcessingContext *ctx = context->priv;
DNNReturnType result;
// have a try run in case that the dnn model resize the frame
result = (ctx->dnn_module->execute_model)(ctx->model, &ctx->output, 1);
if (result != DNN_SUCCESS){
av_log(ctx, AV_LOG_ERROR, "failed to execute model\n");
return AVERROR(EIO);
}
outlink->w = ctx->output.width;
outlink->h = ctx->output.height;
return 0;
}
static int copy_from_frame_to_dnn(DNNData *dnn_input, const AVFrame *frame)
{
switch (frame->format) {
case AV_PIX_FMT_RGB24:
case AV_PIX_FMT_BGR24:
if (dnn_input->dt == DNN_FLOAT) {
float *dnn_input_data = dnn_input->data;
for (int i = 0; i < frame->height; i++) {
for(int j = 0; j < frame->width * 3; j++) {
int k = i * frame->linesize[0] + j;
int t = i * frame->width * 3 + j;
dnn_input_data[t] = frame->data[0][k] / 255.0f;
}
}
} else {
uint8_t *dnn_input_data = dnn_input->data;
av_assert0(dnn_input->dt == DNN_UINT8);
for (int i = 0; i < frame->height; i++) {
for(int j = 0; j < frame->width * 3; j++) {
int k = i * frame->linesize[0] + j;
int t = i * frame->width * 3 + j;
dnn_input_data[t] = frame->data[0][k];
}
}
}
return 0;
case AV_PIX_FMT_GRAY8:
{
uint8_t *dnn_input_data = dnn_input->data;
av_assert0(dnn_input->dt == DNN_UINT8);
for (int i = 0; i < frame->height; i++) {
for(int j = 0; j < frame->width; j++) {
int k = i * frame->linesize[0] + j;
int t = i * frame->width + j;
dnn_input_data[t] = frame->data[0][k];
}
}
}
return 0;
case AV_PIX_FMT_GRAYF32:
{
float *dnn_input_data = dnn_input->data;
av_assert0(dnn_input->dt == DNN_FLOAT);
for (int i = 0; i < frame->height; i++) {
for(int j = 0; j < frame->width; j++) {
int k = i * frame->linesize[0] + j * sizeof(float);
int t = i * frame->width + j;
dnn_input_data[t] = *(float*)(frame->data[0] + k);
}
}
}
return 0;
default:
return AVERROR(EIO);
}
return 0;
}
static int copy_from_dnn_to_frame(AVFrame *frame, const DNNData *dnn_output)
{
switch (frame->format) {
case AV_PIX_FMT_RGB24:
case AV_PIX_FMT_BGR24:
if (dnn_output->dt == DNN_FLOAT) {
float *dnn_output_data = dnn_output->data;
for (int i = 0; i < frame->height; i++) {
for(int j = 0; j < frame->width * 3; j++) {
int k = i * frame->linesize[0] + j;
int t = i * frame->width * 3 + j;
frame->data[0][k] = av_clip_uintp2((int)(dnn_output_data[t] * 255.0f), 8);
}
}
} else {
uint8_t *dnn_output_data = dnn_output->data;
av_assert0(dnn_output->dt == DNN_UINT8);
for (int i = 0; i < frame->height; i++) {
for(int j = 0; j < frame->width * 3; j++) {
int k = i * frame->linesize[0] + j;
int t = i * frame->width * 3 + j;
frame->data[0][k] = dnn_output_data[t];
}
}
}
return 0;
case AV_PIX_FMT_GRAY8:
{
uint8_t *dnn_output_data = dnn_output->data;
av_assert0(dnn_output->dt == DNN_UINT8);
for (int i = 0; i < frame->height; i++) {
for(int j = 0; j < frame->width; j++) {
int k = i * frame->linesize[0] + j;
int t = i * frame->width + j;
frame->data[0][k] = dnn_output_data[t];
}
}
}
return 0;
case AV_PIX_FMT_GRAYF32:
{
float *dnn_output_data = dnn_output->data;
av_assert0(dnn_output->dt == DNN_FLOAT);
for (int i = 0; i < frame->height; i++) {
for(int j = 0; j < frame->width; j++) {
int k = i * frame->linesize[0] + j * sizeof(float);
int t = i * frame->width + j;
*(float*)(frame->data[0] + k) = dnn_output_data[t];
}
}
}
return 0;
default:
return AVERROR(EIO);
}
return 0;
}
static int filter_frame(AVFilterLink *inlink, AVFrame *in)
{
AVFilterContext *context = inlink->dst;
AVFilterLink *outlink = context->outputs[0];
DnnProcessingContext *ctx = context->priv;
DNNReturnType dnn_result;
AVFrame *out;
copy_from_frame_to_dnn(&ctx->input, in);
dnn_result = (ctx->dnn_module->execute_model)(ctx->model, &ctx->output, 1);
if (dnn_result != DNN_SUCCESS){
av_log(ctx, AV_LOG_ERROR, "failed to execute model\n");
av_frame_free(&in);
return AVERROR(EIO);
}
out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
if (!out) {
av_frame_free(&in);
return AVERROR(ENOMEM);
}
av_frame_copy_props(out, in);
copy_from_dnn_to_frame(out, &ctx->output);
av_frame_free(&in);
return ff_filter_frame(outlink, out);
}
static av_cold void uninit(AVFilterContext *ctx)
{
DnnProcessingContext *context = ctx->priv;
if (context->dnn_module)
(context->dnn_module->free_model)(&context->model);
av_freep(&context->dnn_module);
}
static const AVFilterPad dnn_processing_inputs[] = {
{
.name = "default",
.type = AVMEDIA_TYPE_VIDEO,
.config_props = config_input,
.filter_frame = filter_frame,
},
{ NULL }
};
static const AVFilterPad dnn_processing_outputs[] = {
{
.name = "default",
.type = AVMEDIA_TYPE_VIDEO,
.config_props = config_output,
},
{ NULL }
};
AVFilter ff_vf_dnn_processing = {
.name = "dnn_processing",
.description = NULL_IF_CONFIG_SMALL("Apply DNN processing filter to the input."),
.priv_size = sizeof(DnnProcessingContext),
.init = init,
.uninit = uninit,
.query_formats = query_formats,
.inputs = dnn_processing_inputs,
.outputs = dnn_processing_outputs,
.priv_class = &dnn_processing_class,
};
|