FFmpeg
vf_sr.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018 Sergey Lavrushkin
3  *
4  * This file is part of FFmpeg.
5  *
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.
10  *
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.
15  *
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
19  */
20 
21 /**
22  * @file
23  * Filter implementing image super-resolution using deep convolutional networks.
24  * https://arxiv.org/abs/1501.00092
25  * https://arxiv.org/abs/1609.05158
26  */
27 
28 #include "avfilter.h"
29 #include "formats.h"
30 #include "internal.h"
31 #include "libavutil/opt.h"
32 #include "libavutil/pixdesc.h"
33 #include "libavformat/avio.h"
34 #include "libswscale/swscale.h"
35 #include "dnn_filter_common.h"
36 
37 typedef struct SRContext {
38  const AVClass *class;
44 } SRContext;
45 
46 #define OFFSET(x) offsetof(SRContext, x)
47 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM
48 static const AVOption sr_options[] = {
49  { "dnn_backend", "DNN backend used for model execution", OFFSET(dnnctx.backend_type), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS, "backend" },
50  { "native", "native backend flag", 0, AV_OPT_TYPE_CONST, { .i64 = 0 }, 0, 0, FLAGS, "backend" },
51 #if (CONFIG_LIBTENSORFLOW == 1)
52  { "tensorflow", "tensorflow backend flag", 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, 0, 0, FLAGS, "backend" },
53 #endif
54  { "scale_factor", "scale factor for SRCNN model", OFFSET(scale_factor), AV_OPT_TYPE_INT, { .i64 = 2 }, 2, 4, FLAGS },
55  { "model", "path to model file specifying network architecture and its parameters", OFFSET(dnnctx.model_filename), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, FLAGS },
56  { "input", "input name of the model", OFFSET(dnnctx.model_inputname), AV_OPT_TYPE_STRING, { .str = "x" }, 0, 0, FLAGS },
57  { "output", "output name of the model", OFFSET(dnnctx.model_outputname), AV_OPT_TYPE_STRING, { .str = "y" }, 0, 0, FLAGS },
58  { NULL }
59 };
60 
62 
64 {
65  SRContext *sr_context = context->priv;
66  return ff_dnn_init(&sr_context->dnnctx, DFT_PROCESS_FRAME, context);
67 }
68 
70 {
74  AVFilterFormats *formats_list;
75 
76  formats_list = ff_make_format_list(pixel_formats);
77  if (!formats_list){
78  av_log(context, AV_LOG_ERROR, "could not create formats list\n");
79  return AVERROR(ENOMEM);
80  }
81 
82  return ff_set_common_formats(context, formats_list);
83 }
84 
85 static int config_output(AVFilterLink *outlink)
86 {
87  AVFilterContext *context = outlink->src;
88  SRContext *ctx = context->priv;
90  AVFilterLink *inlink = context->inputs[0];
91  int out_width, out_height;
92 
93  // have a try run in case that the dnn model resize the frame
94  result = ff_dnn_get_output(&ctx->dnnctx, inlink->w, inlink->h, &out_width, &out_height);
95  if (result != DNN_SUCCESS) {
96  av_log(ctx, AV_LOG_ERROR, "could not get output from the model\n");
97  return AVERROR(EIO);
98  }
99 
100  if (inlink->w != out_width || inlink->h != out_height) {
101  //espcn
102  outlink->w = out_width;
103  outlink->h = out_height;
104  if (inlink->format != AV_PIX_FMT_GRAY8){
106  int sws_src_h = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
107  int sws_src_w = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
108  int sws_dst_h = AV_CEIL_RSHIFT(outlink->h, desc->log2_chroma_h);
109  int sws_dst_w = AV_CEIL_RSHIFT(outlink->w, desc->log2_chroma_w);
110  ctx->sws_uv_scale = sws_getContext(sws_src_w, sws_src_h, AV_PIX_FMT_GRAY8,
111  sws_dst_w, sws_dst_h, AV_PIX_FMT_GRAY8,
113  ctx->sws_uv_height = sws_src_h;
114  }
115  } else {
116  //srcnn
117  outlink->w = out_width * ctx->scale_factor;
118  outlink->h = out_height * ctx->scale_factor;
119  ctx->sws_pre_scale = sws_getContext(inlink->w, inlink->h, inlink->format,
120  outlink->w, outlink->h, outlink->format,
122  }
123 
124  return 0;
125 }
126 
128 {
130  SRContext *ctx = context->priv;
131  AVFilterLink *outlink = context->outputs[0];
132  AVFrame *out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
133  DNNReturnType dnn_result;
134 
135  if (!out){
136  av_log(context, AV_LOG_ERROR, "could not allocate memory for output frame\n");
137  av_frame_free(&in);
138  return AVERROR(ENOMEM);
139  }
141 
142  if (ctx->sws_pre_scale) {
143  sws_scale(ctx->sws_pre_scale,
144  (const uint8_t **)in->data, in->linesize, 0, in->height,
145  out->data, out->linesize);
146  dnn_result = ff_dnn_execute_model(&ctx->dnnctx, out, out);
147  } else {
148  dnn_result = ff_dnn_execute_model(&ctx->dnnctx, in, out);
149  }
150 
151  if (dnn_result != DNN_SUCCESS){
152  av_log(ctx, AV_LOG_ERROR, "failed to execute loaded model\n");
153  av_frame_free(&in);
154  av_frame_free(&out);
155  return AVERROR(EIO);
156  }
157 
158  if (ctx->sws_uv_scale) {
159  sws_scale(ctx->sws_uv_scale, (const uint8_t **)(in->data + 1), in->linesize + 1,
160  0, ctx->sws_uv_height, out->data + 1, out->linesize + 1);
161  sws_scale(ctx->sws_uv_scale, (const uint8_t **)(in->data + 2), in->linesize + 2,
162  0, ctx->sws_uv_height, out->data + 2, out->linesize + 2);
163  }
164 
165  av_frame_free(&in);
166  return ff_filter_frame(outlink, out);
167 }
168 
170 {
171  SRContext *sr_context = context->priv;
172 
173  ff_dnn_uninit(&sr_context->dnnctx);
174  sws_freeContext(sr_context->sws_uv_scale);
175  sws_freeContext(sr_context->sws_pre_scale);
176 }
177 
178 static const AVFilterPad sr_inputs[] = {
179  {
180  .name = "default",
181  .type = AVMEDIA_TYPE_VIDEO,
182  .filter_frame = filter_frame,
183  },
184  { NULL }
185 };
186 
187 static const AVFilterPad sr_outputs[] = {
188  {
189  .name = "default",
190  .config_props = config_output,
191  .type = AVMEDIA_TYPE_VIDEO,
192  },
193  { NULL }
194 };
195 
197  .name = "sr",
198  .description = NULL_IF_CONFIG_SMALL("Apply DNN-based image super resolution to the input."),
199  .priv_size = sizeof(SRContext),
200  .init = init,
201  .uninit = uninit,
203  .inputs = sr_inputs,
204  .outputs = sr_outputs,
205  .priv_class = &sr_class,
206 };
ff_get_video_buffer
AVFrame * ff_get_video_buffer(AVFilterLink *link, int w, int h)
Request a picture buffer with a specific set of permissions.
Definition: video.c:99
AVPixelFormat
AVPixelFormat
Pixel format.
Definition: pixfmt.h:64
SRContext::dnnctx
DnnContext dnnctx
Definition: vf_sr.c:39
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
opt.h
ff_make_format_list
AVFilterFormats * ff_make_format_list(const int *fmts)
Create a list of supported formats.
Definition: formats.c:286
ff_dnn_get_output
DNNReturnType ff_dnn_get_output(DnnContext *ctx, int input_width, int input_height, int *output_width, int *output_height)
Definition: dnn_filter_common.c:85
sr_outputs
static const AVFilterPad sr_outputs[]
Definition: vf_sr.c:187
out
FILE * out
Definition: movenc.c:54
sr_options
static const AVOption sr_options[]
Definition: vf_sr.c:48
SRContext::sws_pre_scale
struct SwsContext * sws_pre_scale
Definition: vf_sr.c:43
ff_filter_frame
int ff_filter_frame(AVFilterLink *link, AVFrame *frame)
Send a frame of data to the next filter.
Definition: avfilter.c:1094
av_pix_fmt_desc_get
const AVPixFmtDescriptor * av_pix_fmt_desc_get(enum AVPixelFormat pix_fmt)
Definition: pixdesc.c:2573
inlink
The exact code depends on how similar the blocks are and how related they are to the and needs to apply these operations to the correct inlink or outlink if there are several Macros are available to factor that when no extra processing is inlink
Definition: filter_design.txt:212
av_frame_free
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
Definition: frame.c:204
AVFrame
This structure describes decoded (raw) audio or video data.
Definition: frame.h:324
pixdesc.h
AVOption
AVOption.
Definition: opt.h:248
sws_scale
int attribute_align_arg sws_scale(struct SwsContext *c, const uint8_t *const srcSlice[], const int srcStride[], int srcSliceY, int srcSliceH, uint8_t *const dst[], const int dstStride[])
swscale wrapper, so we don't need to export the SwsContext.
Definition: swscale.c:745
AVFilter::name
const char * name
Filter name.
Definition: avfilter.h:149
SRContext
Definition: af_acrusher.c:35
dnn_filter_common.h
AVFilterFormats
A list of supported formats for one end of a filter link.
Definition: formats.h:65
formats.h
uninit
static av_cold void uninit(AVFilterContext *context)
Definition: vf_sr.c:169
DnnContext
Definition: dnn_filter_common.h:29
DNN_SUCCESS
@ DNN_SUCCESS
Definition: dnn_interface.h:33
AVFilterPad
A filter pad used for either input or output.
Definition: internal.h:54
query_formats
static int query_formats(AVFilterContext *context)
Definition: vf_sr.c:69
AV_LOG_ERROR
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:194
av_cold
#define av_cold
Definition: attributes.h:90
inputs
these buffered frames must be flushed immediately if a new input produces new the filter must not call request_frame to get more It must just process the frame or queue it The task of requesting more frames is left to the filter s request_frame method or the application If a filter has several inputs
Definition: filter_design.txt:243
ff_set_common_formats
int ff_set_common_formats(AVFilterContext *ctx, AVFilterFormats *formats)
A helper for query_formats() which sets all links to the same list of formats.
Definition: formats.c:587
init
static av_cold int init(AVFilterContext *context)
Definition: vf_sr.c:63
AV_CEIL_RSHIFT
#define AV_CEIL_RSHIFT(a, b)
Definition: common.h:58
FLAGS
#define FLAGS
Definition: vf_sr.c:47
DNNReturnType
DNNReturnType
Definition: dnn_interface.h:33
outputs
static const AVFilterPad outputs[]
Definition: af_acontrast.c:203
ctx
AVFormatContext * ctx
Definition: movenc.c:48
AVFILTER_DEFINE_CLASS
AVFILTER_DEFINE_CLASS(sr)
AV_PIX_FMT_YUV420P
@ AV_PIX_FMT_YUV420P
planar YUV 4:2:0, 12bpp, (1 Cr & Cb sample per 2x2 Y samples)
Definition: pixfmt.h:66
context
it s the only field you need to keep assuming you have a context There is some magic you don t need to care about around this just let it vf default minimum maximum flags name is the option keep it simple and lowercase description are in without and describe what they for example set the foo of the bar offset is the offset of the field in your context
Definition: writing_filters.txt:91
OFFSET
#define OFFSET(x)
Definition: vf_sr.c:46
AVClass
Describe the class of an AVClass context structure.
Definition: log.h:67
result
and forward the result(frame or status change) to the corresponding input. If nothing is possible
NULL
#define NULL
Definition: coverity.c:32
av_frame_copy_props
int av_frame_copy_props(AVFrame *dst, const AVFrame *src)
Copy only "metadata" fields from src to dst.
Definition: frame.c:659
ff_vf_sr
AVFilter ff_vf_sr
Definition: vf_sr.c:196
AV_PIX_FMT_GRAY8
@ AV_PIX_FMT_GRAY8
Y , 8bpp.
Definition: pixfmt.h:74
SRContext::sws_uv_scale
struct SwsContext * sws_uv_scale
Definition: vf_sr.c:41
SRContext::sws_uv_height
int sws_uv_height
Definition: vf_sr.c:42
NULL_IF_CONFIG_SMALL
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification.
Definition: internal.h:117
sws_getContext
struct SwsContext * sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat, int dstW, int dstH, enum AVPixelFormat dstFormat, int flags, SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
Allocate and return an SwsContext.
Definition: utils.c:1917
avio.h
internal.h
in
uint8_t pi<< 24) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_U8, uint8_t,(*(const uint8_t *) pi - 0x80) *(1.0f/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_U8, uint8_t,(*(const uint8_t *) pi - 0x80) *(1.0/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S16, int16_t,(*(const int16_t *) pi >> 8)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S16, int16_t, *(const int16_t *) pi *(1.0f/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S16, int16_t, *(const int16_t *) pi *(1.0/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S32, int32_t,(*(const int32_t *) pi >> 24)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S32, int32_t, *(const int32_t *) pi *(1.0f/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S32, int32_t, *(const int32_t *) pi *(1.0/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_FLT, float, av_clip_uint8(lrintf(*(const float *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_FLT, float, av_clip_int16(lrintf(*(const float *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_FLT, float, av_clipl_int32(llrintf(*(const float *) pi *(1U<< 31)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_DBL, double, av_clip_uint8(lrint(*(const double *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_DBL, double, av_clip_int16(lrint(*(const double *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_DBL, double, av_clipl_int32(llrint(*(const double *) pi *(1U<< 31)))) #define SET_CONV_FUNC_GROUP(ofmt, ifmt) static void set_generic_function(AudioConvert *ac) { } void ff_audio_convert_free(AudioConvert **ac) { if(! *ac) return;ff_dither_free(&(*ac) ->dc);av_freep(ac);} AudioConvert *ff_audio_convert_alloc(AVAudioResampleContext *avr, enum AVSampleFormat out_fmt, enum AVSampleFormat in_fmt, int channels, int sample_rate, int apply_map) { AudioConvert *ac;int in_planar, out_planar;ac=av_mallocz(sizeof(*ac));if(!ac) return NULL;ac->avr=avr;ac->out_fmt=out_fmt;ac->in_fmt=in_fmt;ac->channels=channels;ac->apply_map=apply_map;if(avr->dither_method !=AV_RESAMPLE_DITHER_NONE &&av_get_packed_sample_fmt(out_fmt)==AV_SAMPLE_FMT_S16 &&av_get_bytes_per_sample(in_fmt) > 2) { ac->dc=ff_dither_alloc(avr, out_fmt, in_fmt, channels, sample_rate, apply_map);if(!ac->dc) { av_free(ac);return NULL;} return ac;} in_planar=ff_sample_fmt_is_planar(in_fmt, channels);out_planar=ff_sample_fmt_is_planar(out_fmt, channels);if(in_planar==out_planar) { ac->func_type=CONV_FUNC_TYPE_FLAT;ac->planes=in_planar ? ac->channels :1;} else if(in_planar) ac->func_type=CONV_FUNC_TYPE_INTERLEAVE;else ac->func_type=CONV_FUNC_TYPE_DEINTERLEAVE;set_generic_function(ac);if(ARCH_AARCH64) ff_audio_convert_init_aarch64(ac);if(ARCH_ARM) ff_audio_convert_init_arm(ac);if(ARCH_X86) ff_audio_convert_init_x86(ac);return ac;} int ff_audio_convert(AudioConvert *ac, AudioData *out, AudioData *in) { int use_generic=1;int len=in->nb_samples;int p;if(ac->dc) { av_log(ac->avr, AV_LOG_TRACE, "%d samples - audio_convert: %s to %s (dithered)\n", len, av_get_sample_fmt_name(ac->in_fmt), av_get_sample_fmt_name(ac->out_fmt));return ff_convert_dither(ac-> in
Definition: audio_convert.c:326
filter_frame
static int filter_frame(AVFilterLink *inlink, AVFrame *in)
Definition: vf_sr.c:127
uint8_t
uint8_t
Definition: audio_convert.c:194
AVFilterPad::name
const char * name
Pad name.
Definition: internal.h:60
AVFilter
Filter definition.
Definition: avfilter.h:145
SRContext::scale_factor
int scale_factor
Definition: vf_sr.c:40
sws_freeContext
void sws_freeContext(struct SwsContext *swsContext)
Free the swscaler context swsContext.
Definition: utils.c:2337
AV_PIX_FMT_NONE
@ AV_PIX_FMT_NONE
Definition: pixfmt.h:65
AV_OPT_TYPE_INT
@ AV_OPT_TYPE_INT
Definition: opt.h:225
avfilter.h
AV_PIX_FMT_YUV444P
@ AV_PIX_FMT_YUV444P
planar YUV 4:4:4, 24bpp, (1 Cr & Cb sample per 1x1 Y samples)
Definition: pixfmt.h:71
AVFilterContext
An instance of a filter.
Definition: avfilter.h:341
desc
const char * desc
Definition: libsvtav1.c:79
AVMEDIA_TYPE_VIDEO
@ AVMEDIA_TYPE_VIDEO
Definition: avutil.h:201
AV_PIX_FMT_YUV422P
@ AV_PIX_FMT_YUV422P
planar YUV 4:2:2, 16bpp, (1 Cr & Cb sample per 2x1 Y samples)
Definition: pixfmt.h:70
AVPixFmtDescriptor
Descriptor that unambiguously describes how the bits of a pixel are stored in the up to 4 data planes...
Definition: pixdesc.h:81
ff_dnn_execute_model
DNNReturnType ff_dnn_execute_model(DnnContext *ctx, AVFrame *in_frame, AVFrame *out_frame)
Definition: dnn_filter_common.c:91
sr_inputs
static const AVFilterPad sr_inputs[]
Definition: vf_sr.c:178
ff_dnn_init
int ff_dnn_init(DnnContext *ctx, DNNFunctionType func_type, AVFilterContext *filter_ctx)
Definition: dnn_filter_common.c:21
AV_PIX_FMT_YUV411P
@ AV_PIX_FMT_YUV411P
planar YUV 4:1:1, 12bpp, (1 Cr & Cb sample per 4x1 Y samples)
Definition: pixfmt.h:73
AV_PIX_FMT_YUV410P
@ AV_PIX_FMT_YUV410P
planar YUV 4:1:0, 9bpp, (1 Cr & Cb sample per 4x4 Y samples)
Definition: pixfmt.h:72
av_log
#define av_log(a,...)
Definition: tableprint_vlc.h:28
ff_dnn_uninit
void ff_dnn_uninit(DnnContext *ctx)
Definition: dnn_filter_common.c:113
AV_OPT_TYPE_STRING
@ AV_OPT_TYPE_STRING
Definition: opt.h:229
SwsContext
Definition: swscale_internal.h:283
AV_OPT_TYPE_CONST
@ AV_OPT_TYPE_CONST
Definition: opt.h:234
DFT_PROCESS_FRAME
@ DFT_PROCESS_FRAME
Definition: dnn_interface.h:53
SWS_BICUBIC
#define SWS_BICUBIC
Definition: swscale.h:60
swscale.h
config_output
static int config_output(AVFilterLink *outlink)
Definition: vf_sr.c:85