Line |
Branch |
Exec |
Source |
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 "internal.h" |
30 |
|
|
#include "video.h" |
31 |
|
|
#include "libavutil/opt.h" |
32 |
|
|
#include "libavutil/pixdesc.h" |
33 |
|
|
#include "libswscale/swscale.h" |
34 |
|
|
#include "dnn_filter_common.h" |
35 |
|
|
|
36 |
|
|
typedef struct SRContext { |
37 |
|
|
const AVClass *class; |
38 |
|
|
DnnContext dnnctx; |
39 |
|
|
int scale_factor; |
40 |
|
|
struct SwsContext *sws_uv_scale; |
41 |
|
|
int sws_uv_height; |
42 |
|
|
struct SwsContext *sws_pre_scale; |
43 |
|
|
} SRContext; |
44 |
|
|
|
45 |
|
|
#define OFFSET(x) offsetof(SRContext, x) |
46 |
|
|
#define FLAGS AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM |
47 |
|
|
static const AVOption sr_options[] = { |
48 |
|
|
{ "dnn_backend", "DNN backend used for model execution", OFFSET(dnnctx.backend_type), AV_OPT_TYPE_INT, { .i64 = 1 }, 0, 1, FLAGS, .unit = "backend" }, |
49 |
|
|
#if (CONFIG_LIBTENSORFLOW == 1) |
50 |
|
|
{ "tensorflow", "tensorflow backend flag", 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, 0, 0, FLAGS, .unit = "backend" }, |
51 |
|
|
#endif |
52 |
|
|
{ "scale_factor", "scale factor for SRCNN model", OFFSET(scale_factor), AV_OPT_TYPE_INT, { .i64 = 2 }, 2, 4, FLAGS }, |
53 |
|
|
{ "model", "path to model file specifying network architecture and its parameters", OFFSET(dnnctx.model_filename), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, FLAGS }, |
54 |
|
|
{ "input", "input name of the model", OFFSET(dnnctx.model_inputname), AV_OPT_TYPE_STRING, { .str = "x" }, 0, 0, FLAGS }, |
55 |
|
|
{ "output", "output name of the model", OFFSET(dnnctx.model_outputnames_string), AV_OPT_TYPE_STRING, { .str = "y" }, 0, 0, FLAGS }, |
56 |
|
|
{ NULL } |
57 |
|
|
}; |
58 |
|
|
|
59 |
|
|
AVFILTER_DEFINE_CLASS(sr); |
60 |
|
|
|
61 |
|
✗ |
static av_cold int init(AVFilterContext *context) |
62 |
|
|
{ |
63 |
|
✗ |
SRContext *sr_context = context->priv; |
64 |
|
✗ |
return ff_dnn_init(&sr_context->dnnctx, DFT_PROCESS_FRAME, context); |
65 |
|
|
} |
66 |
|
|
|
67 |
|
|
static const enum AVPixelFormat pixel_formats[] = { |
68 |
|
|
AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P, |
69 |
|
|
AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_GRAY8, |
70 |
|
|
AV_PIX_FMT_NONE |
71 |
|
|
}; |
72 |
|
|
|
73 |
|
✗ |
static int config_output(AVFilterLink *outlink) |
74 |
|
|
{ |
75 |
|
✗ |
AVFilterContext *context = outlink->src; |
76 |
|
✗ |
SRContext *ctx = context->priv; |
77 |
|
|
int result; |
78 |
|
✗ |
AVFilterLink *inlink = context->inputs[0]; |
79 |
|
|
int out_width, out_height; |
80 |
|
|
|
81 |
|
|
// have a try run in case that the dnn model resize the frame |
82 |
|
✗ |
result = ff_dnn_get_output(&ctx->dnnctx, inlink->w, inlink->h, &out_width, &out_height); |
83 |
|
✗ |
if (result != 0) { |
84 |
|
✗ |
av_log(ctx, AV_LOG_ERROR, "could not get output from the model\n"); |
85 |
|
✗ |
return result; |
86 |
|
|
} |
87 |
|
|
|
88 |
|
✗ |
if (inlink->w != out_width || inlink->h != out_height) { |
89 |
|
|
//espcn |
90 |
|
✗ |
outlink->w = out_width; |
91 |
|
✗ |
outlink->h = out_height; |
92 |
|
✗ |
if (inlink->format != AV_PIX_FMT_GRAY8){ |
93 |
|
✗ |
const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format); |
94 |
|
✗ |
int sws_src_h = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h); |
95 |
|
✗ |
int sws_src_w = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w); |
96 |
|
✗ |
int sws_dst_h = AV_CEIL_RSHIFT(outlink->h, desc->log2_chroma_h); |
97 |
|
✗ |
int sws_dst_w = AV_CEIL_RSHIFT(outlink->w, desc->log2_chroma_w); |
98 |
|
✗ |
ctx->sws_uv_scale = sws_getContext(sws_src_w, sws_src_h, AV_PIX_FMT_GRAY8, |
99 |
|
|
sws_dst_w, sws_dst_h, AV_PIX_FMT_GRAY8, |
100 |
|
|
SWS_BICUBIC, NULL, NULL, NULL); |
101 |
|
✗ |
ctx->sws_uv_height = sws_src_h; |
102 |
|
|
} |
103 |
|
|
} else { |
104 |
|
|
//srcnn |
105 |
|
✗ |
outlink->w = out_width * ctx->scale_factor; |
106 |
|
✗ |
outlink->h = out_height * ctx->scale_factor; |
107 |
|
✗ |
ctx->sws_pre_scale = sws_getContext(inlink->w, inlink->h, inlink->format, |
108 |
|
✗ |
outlink->w, outlink->h, outlink->format, |
109 |
|
|
SWS_BICUBIC, NULL, NULL, NULL); |
110 |
|
|
} |
111 |
|
|
|
112 |
|
✗ |
return 0; |
113 |
|
|
} |
114 |
|
|
|
115 |
|
✗ |
static int filter_frame(AVFilterLink *inlink, AVFrame *in) |
116 |
|
|
{ |
117 |
|
✗ |
DNNAsyncStatusType async_state = 0; |
118 |
|
✗ |
AVFilterContext *context = inlink->dst; |
119 |
|
✗ |
SRContext *ctx = context->priv; |
120 |
|
✗ |
AVFilterLink *outlink = context->outputs[0]; |
121 |
|
✗ |
AVFrame *out = ff_get_video_buffer(outlink, outlink->w, outlink->h); |
122 |
|
|
int dnn_result; |
123 |
|
|
|
124 |
|
✗ |
if (!out){ |
125 |
|
✗ |
av_log(context, AV_LOG_ERROR, "could not allocate memory for output frame\n"); |
126 |
|
✗ |
av_frame_free(&in); |
127 |
|
✗ |
return AVERROR(ENOMEM); |
128 |
|
|
} |
129 |
|
✗ |
av_frame_copy_props(out, in); |
130 |
|
|
|
131 |
|
✗ |
if (ctx->sws_pre_scale) { |
132 |
|
✗ |
sws_scale(ctx->sws_pre_scale, |
133 |
|
✗ |
(const uint8_t **)in->data, in->linesize, 0, in->height, |
134 |
|
✗ |
out->data, out->linesize); |
135 |
|
✗ |
dnn_result = ff_dnn_execute_model(&ctx->dnnctx, out, out); |
136 |
|
|
} else { |
137 |
|
✗ |
dnn_result = ff_dnn_execute_model(&ctx->dnnctx, in, out); |
138 |
|
|
} |
139 |
|
|
|
140 |
|
✗ |
if (dnn_result != 0){ |
141 |
|
✗ |
av_log(ctx, AV_LOG_ERROR, "failed to execute loaded model\n"); |
142 |
|
✗ |
av_frame_free(&in); |
143 |
|
✗ |
av_frame_free(&out); |
144 |
|
✗ |
return dnn_result; |
145 |
|
|
} |
146 |
|
|
|
147 |
|
|
do { |
148 |
|
✗ |
async_state = ff_dnn_get_result(&ctx->dnnctx, &in, &out); |
149 |
|
✗ |
} while (async_state == DAST_NOT_READY); |
150 |
|
|
|
151 |
|
✗ |
if (async_state != DAST_SUCCESS) |
152 |
|
✗ |
return AVERROR(EINVAL); |
153 |
|
|
|
154 |
|
✗ |
if (ctx->sws_uv_scale) { |
155 |
|
✗ |
sws_scale(ctx->sws_uv_scale, (const uint8_t **)(in->data + 1), in->linesize + 1, |
156 |
|
✗ |
0, ctx->sws_uv_height, out->data + 1, out->linesize + 1); |
157 |
|
✗ |
sws_scale(ctx->sws_uv_scale, (const uint8_t **)(in->data + 2), in->linesize + 2, |
158 |
|
✗ |
0, ctx->sws_uv_height, out->data + 2, out->linesize + 2); |
159 |
|
|
} |
160 |
|
✗ |
if (in != out) { |
161 |
|
✗ |
av_frame_free(&in); |
162 |
|
|
} |
163 |
|
✗ |
return ff_filter_frame(outlink, out); |
164 |
|
|
} |
165 |
|
|
|
166 |
|
✗ |
static av_cold void uninit(AVFilterContext *context) |
167 |
|
|
{ |
168 |
|
✗ |
SRContext *sr_context = context->priv; |
169 |
|
|
|
170 |
|
✗ |
ff_dnn_uninit(&sr_context->dnnctx); |
171 |
|
✗ |
sws_freeContext(sr_context->sws_uv_scale); |
172 |
|
✗ |
sws_freeContext(sr_context->sws_pre_scale); |
173 |
|
✗ |
} |
174 |
|
|
|
175 |
|
|
static const AVFilterPad sr_inputs[] = { |
176 |
|
|
{ |
177 |
|
|
.name = "default", |
178 |
|
|
.type = AVMEDIA_TYPE_VIDEO, |
179 |
|
|
.filter_frame = filter_frame, |
180 |
|
|
}, |
181 |
|
|
}; |
182 |
|
|
|
183 |
|
|
static const AVFilterPad sr_outputs[] = { |
184 |
|
|
{ |
185 |
|
|
.name = "default", |
186 |
|
|
.config_props = config_output, |
187 |
|
|
.type = AVMEDIA_TYPE_VIDEO, |
188 |
|
|
}, |
189 |
|
|
}; |
190 |
|
|
|
191 |
|
|
const AVFilter ff_vf_sr = { |
192 |
|
|
.name = "sr", |
193 |
|
|
.description = NULL_IF_CONFIG_SMALL("Apply DNN-based image super resolution to the input."), |
194 |
|
|
.priv_size = sizeof(SRContext), |
195 |
|
|
.init = init, |
196 |
|
|
.uninit = uninit, |
197 |
|
|
FILTER_INPUTS(sr_inputs), |
198 |
|
|
FILTER_OUTPUTS(sr_outputs), |
199 |
|
|
FILTER_PIXFMTS_ARRAY(pixel_formats), |
200 |
|
|
.priv_class = &sr_class, |
201 |
|
|
}; |
202 |
|
|
|