FFmpeg
vf_derain.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2019 Xuewei Meng
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 derain filter using deep convolutional networks.
24  * http://openaccess.thecvf.com/content_ECCV_2018/html/Xia_Li_Recurrent_Squeeze-and-Excitation_Context_ECCV_2018_paper.html
25  */
26 
27 #include "libavformat/avio.h"
28 #include "libavutil/opt.h"
29 #include "avfilter.h"
30 #include "dnn_interface.h"
31 #include "formats.h"
32 #include "internal.h"
33 
34 typedef struct DRContext {
35  const AVClass *class;
36 
44 } DRContext;
45 
46 #define CLIP(x, min, max) (x < min ? min : (x > max ? max : x))
47 #define OFFSET(x) offsetof(DRContext, x)
48 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM
49 static const AVOption derain_options[] = {
50  { "filter_type", "filter type(derain/dehaze)", OFFSET(filter_type), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS, "type" },
51  { "derain", "derain filter flag", 0, AV_OPT_TYPE_CONST, { .i64 = 0 }, 0, 0, FLAGS, "type" },
52  { "dehaze", "dehaze filter flag", 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, 0, 0, FLAGS, "type" },
53  { "dnn_backend", "DNN backend", OFFSET(backend_type), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS, "backend" },
54  { "native", "native backend flag", 0, AV_OPT_TYPE_CONST, { .i64 = 0 }, 0, 0, FLAGS, "backend" },
55 #if (CONFIG_LIBTENSORFLOW == 1)
56  { "tensorflow", "tensorflow backend flag", 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, 0, 0, FLAGS, "backend" },
57 #endif
58  { "model", "path to model file", OFFSET(model_filename), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, FLAGS },
59  { NULL }
60 };
61 
62 AVFILTER_DEFINE_CLASS(derain);
63 
65 {
67  const enum AVPixelFormat pixel_fmts[] = {
70  };
71 
72  formats = ff_make_format_list(pixel_fmts);
73 
74  return ff_set_common_formats(ctx, formats);
75 }
76 
78 {
79  AVFilterContext *ctx = inlink->dst;
80  DRContext *dr_context = ctx->priv;
81  const char *model_output_name = "y";
83 
84  dr_context->input.width = inlink->w;
85  dr_context->input.height = inlink->h;
86  dr_context->input.channels = 3;
87 
88  result = (dr_context->model->set_input_output)(dr_context->model->model, &dr_context->input, "x", &model_output_name, 1);
89  if (result != DNN_SUCCESS) {
90  av_log(ctx, AV_LOG_ERROR, "could not set input and output for the model\n");
91  return AVERROR(EIO);
92  }
93 
94  return 0;
95 }
96 
98 {
99  AVFilterContext *ctx = inlink->dst;
100  AVFilterLink *outlink = ctx->outputs[0];
101  DRContext *dr_context = ctx->priv;
102  DNNReturnType dnn_result;
103  int pad_size;
104 
105  AVFrame *out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
106  if (!out) {
107  av_log(ctx, AV_LOG_ERROR, "could not allocate memory for output frame\n");
108  av_frame_free(&in);
109  return AVERROR(ENOMEM);
110  }
111 
112  av_frame_copy_props(out, in);
113 
114  for (int i = 0; i < in->height; i++){
115  for(int j = 0; j < in->width * 3; j++){
116  int k = i * in->linesize[0] + j;
117  int t = i * in->width * 3 + j;
118  ((float *)dr_context->input.data)[t] = in->data[0][k] / 255.0;
119  }
120  }
121 
122  dnn_result = (dr_context->dnn_module->execute_model)(dr_context->model, &dr_context->output, 1);
123  if (dnn_result != DNN_SUCCESS){
124  av_log(ctx, AV_LOG_ERROR, "failed to execute model\n");
125  return AVERROR(EIO);
126  }
127 
128  out->height = dr_context->output.height;
129  out->width = dr_context->output.width;
130  outlink->h = dr_context->output.height;
131  outlink->w = dr_context->output.width;
132  pad_size = (in->height - out->height) >> 1;
133 
134  for (int i = 0; i < out->height; i++){
135  for(int j = 0; j < out->width * 3; j++){
136  int k = i * out->linesize[0] + j;
137  int t = i * out->width * 3 + j;
138 
139  int t_in = (i + pad_size) * in->width * 3 + j + pad_size * 3;
140  out->data[0][k] = CLIP((int)((((float *)dr_context->input.data)[t_in] - ((float *)dr_context->output.data)[t]) * 255), 0, 255);
141  }
142  }
143 
144  av_frame_free(&in);
145 
146  return ff_filter_frame(outlink, out);
147 }
148 
150 {
151  DRContext *dr_context = ctx->priv;
152 
153  dr_context->input.dt = DNN_FLOAT;
154  dr_context->dnn_module = ff_get_dnn_module(dr_context->backend_type);
155  if (!dr_context->dnn_module) {
156  av_log(ctx, AV_LOG_ERROR, "could not create DNN module for requested backend\n");
157  return AVERROR(ENOMEM);
158  }
159  if (!dr_context->model_filename) {
160  av_log(ctx, AV_LOG_ERROR, "model file for network is not specified\n");
161  return AVERROR(EINVAL);
162  }
163  if (!dr_context->dnn_module->load_model) {
164  av_log(ctx, AV_LOG_ERROR, "load_model for network is not specified\n");
165  return AVERROR(EINVAL);
166  }
167 
168  dr_context->model = (dr_context->dnn_module->load_model)(dr_context->model_filename);
169  if (!dr_context->model) {
170  av_log(ctx, AV_LOG_ERROR, "could not load DNN model\n");
171  return AVERROR(EINVAL);
172  }
173 
174  return 0;
175 }
176 
178 {
179  DRContext *dr_context = ctx->priv;
180 
181  if (dr_context->dnn_module) {
182  (dr_context->dnn_module->free_model)(&dr_context->model);
183  av_freep(&dr_context->dnn_module);
184  }
185 }
186 
187 static const AVFilterPad derain_inputs[] = {
188  {
189  .name = "default",
190  .type = AVMEDIA_TYPE_VIDEO,
191  .config_props = config_inputs,
192  .filter_frame = filter_frame,
193  },
194  { NULL }
195 };
196 
197 static const AVFilterPad derain_outputs[] = {
198  {
199  .name = "default",
200  .type = AVMEDIA_TYPE_VIDEO,
201  },
202  { NULL }
203 };
204 
206  .name = "derain",
207  .description = NULL_IF_CONFIG_SMALL("Apply derain filter to the input."),
208  .priv_size = sizeof(DRContext),
209  .init = init,
210  .uninit = uninit,
212  .inputs = derain_inputs,
213  .outputs = derain_outputs,
214  .priv_class = &derain_class,
216 };
void * model
Definition: dnn_interface.h:45
#define NULL
Definition: coverity.c:32
Buffered I/O operations.
This structure describes decoded (raw) audio or video data.
Definition: frame.h:295
AVOption.
Definition: opt.h:246
void(* free_model)(DNNModel **model)
Definition: dnn_interface.h:61
Main libavfilter public API header.
packed RGB 8:8:8, 24bpp, RGBRGB...
Definition: pixfmt.h:68
static const AVOption derain_options[]
Definition: vf_derain.c:49
int channels
Definition: dnn_interface.h:40
static int config_inputs(AVFilterLink *inlink)
Definition: vf_derain.c:77
DNNBackendType backend_type
Definition: vf_derain.c:39
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
AVFilterFormats * ff_make_format_list(const int *fmts)
Create a list of supported formats.
Definition: formats.c:283
#define AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC
Some filters support a generic "enable" expression option that can be used to enable or disable a fil...
Definition: avfilter.h:125
const char * name
Pad name.
Definition: internal.h:60
DNNModel *(* load_model)(const char *model_filename)
Definition: dnn_interface.h:57
int ff_filter_frame(AVFilterLink *link, AVFrame *frame)
Send a frame of data to the next filter.
Definition: avfilter.c:1093
#define av_cold
Definition: attributes.h:82
char * model_filename
Definition: vf_derain.c:38
AVOptions.
DNNData input
Definition: vf_derain.c:42
AVFILTER_DEFINE_CLASS(derain)
DNNBackendType
Definition: dnn_interface.h:33
#define av_log(a,...)
A filter pad used for either input or output.
Definition: internal.h:54
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:259
int width
Definition: frame.h:353
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:176
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:569
int height
Definition: dnn_interface.h:40
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
Definition: frame.c:202
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification. ...
Definition: internal.h:186
void * data
Definition: dnn_interface.h:38
void * priv
private data for use by the filter
Definition: avfilter.h:353
static av_cold int init(AVFilterContext *ctx)
Definition: vf_derain.c:149
static av_cold void uninit(AVFilterContext *ctx)
Definition: vf_derain.c:177
DNNData output
Definition: vf_derain.c:43
AVFilter ff_vf_derain
Definition: vf_derain.c:205
DNNModel * model
Definition: vf_derain.c:41
AVFormatContext * ctx
Definition: movenc.c:48
int filter_type
Definition: vf_derain.c:37
DNN inference engine interface.
DNNReturnType
Definition: dnn_interface.h:31
static const AVFilterPad outputs[]
Definition: af_acontrast.c:203
#define OFFSET(x)
Definition: vf_derain.c:47
#define CLIP(x, min, max)
Definition: vf_derain.c:46
int linesize[AV_NUM_DATA_POINTERS]
For video, size in bytes of each picture line.
Definition: frame.h:326
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
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
Describe the class of an AVClass context structure.
Definition: log.h:67
Filter definition.
Definition: avfilter.h:144
const char * name
Filter name.
Definition: avfilter.h:148
AVFilterLink ** outputs
array of pointers to output links
Definition: avfilter.h:350
DNNModule * ff_get_dnn_module(DNNBackendType backend_type)
Definition: dnn_interface.c:31
#define flags(name, subs,...)
Definition: cbs_av1.c:561
DNNReturnType(* execute_model)(const DNNModel *model, DNNData *outputs, uint32_t nb_output)
Definition: dnn_interface.h:59
static const AVFilterPad derain_inputs[]
Definition: vf_derain.c:187
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
Definition: frame.h:309
static int filter_frame(AVFilterLink *inlink, AVFrame *in)
Definition: vf_derain.c:97
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
#define FLAGS
Definition: vf_derain.c:48
static int query_formats(AVFilterContext *ctx)
Definition: vf_derain.c:64
A list of supported formats for one end of a filter link.
Definition: formats.h:64
An instance of a filter.
Definition: avfilter.h:338
DNNModule * dnn_module
Definition: vf_derain.c:40
and forward the result(frame or status change) to the corresponding input.If nothing is possible
int height
Definition: frame.h:353
FILE * out
Definition: movenc.c:54
#define av_freep(p)
DNNReturnType(* set_input_output)(void *model, DNNData *input, const char *input_name, const char **output_names, uint32_t nb_output)
Definition: dnn_interface.h:51
formats
Definition: signature.h:48
internal API functions
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
DNNDataType dt
Definition: dnn_interface.h:39
AVPixelFormat
Pixel format.
Definition: pixfmt.h:64
static const AVFilterPad derain_outputs[]
Definition: vf_derain.c:197
int av_frame_copy_props(AVFrame *dst, const AVFrame *src)
Copy only "metadata" fields from src to dst.
Definition: frame.c:654