2 * Copyright (c) 2018 Sergey Lavrushkin
4 * This file is part of FFmpeg.
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.
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.
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
23 * Filter implementing image super-resolution using deep convolutional networks.
24 * https://arxiv.org/abs/1501.00092
25 * https://arxiv.org/abs/1609.05158
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"
37 typedef struct SRContext {
41 struct SwsContext *sws_uv_scale;
43 struct SwsContext *sws_pre_scale;
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" },
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 },
61 AVFILTER_DEFINE_CLASS(sr);
63 static av_cold int init(AVFilterContext *context)
65 SRContext *sr_context = context->priv;
66 return ff_dnn_init(&sr_context->dnnctx, context);
69 static int query_formats(AVFilterContext *context)
71 const enum AVPixelFormat pixel_formats[] = {AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P,
72 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_GRAY8,
74 AVFilterFormats *formats_list;
76 formats_list = ff_make_format_list(pixel_formats);
78 av_log(context, AV_LOG_ERROR, "could not create formats list\n");
79 return AVERROR(ENOMEM);
82 return ff_set_common_formats(context, formats_list);
85 static int config_output(AVFilterLink *outlink)
87 AVFilterContext *context = outlink->src;
88 SRContext *ctx = context->priv;
90 AVFilterLink *inlink = context->inputs[0];
91 int out_width, out_height;
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");
100 if (inlink->w != out_width || inlink->h != out_height) {
102 outlink->w = out_width;
103 outlink->h = out_height;
104 if (inlink->format != AV_PIX_FMT_GRAY8){
105 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
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,
112 SWS_BICUBIC, NULL, NULL, NULL);
113 ctx->sws_uv_height = sws_src_h;
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,
121 SWS_BICUBIC, NULL, NULL, NULL);
127 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
129 AVFilterContext *context = inlink->dst;
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;
136 av_log(context, AV_LOG_ERROR, "could not allocate memory for output frame\n");
138 return AVERROR(ENOMEM);
140 av_frame_copy_props(out, in);
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);
148 dnn_result = ff_dnn_execute_model(&ctx->dnnctx, in, out);
151 if (dnn_result != DNN_SUCCESS){
152 av_log(ctx, AV_LOG_ERROR, "failed to execute loaded model\n");
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);
166 return ff_filter_frame(outlink, out);
169 static av_cold void uninit(AVFilterContext *context)
171 SRContext *sr_context = context->priv;
173 ff_dnn_uninit(&sr_context->dnnctx);
174 sws_freeContext(sr_context->sws_uv_scale);
175 sws_freeContext(sr_context->sws_pre_scale);
178 static const AVFilterPad sr_inputs[] = {
181 .type = AVMEDIA_TYPE_VIDEO,
182 .filter_frame = filter_frame,
187 static const AVFilterPad sr_outputs[] = {
190 .config_props = config_output,
191 .type = AVMEDIA_TYPE_VIDEO,
196 AVFilter ff_vf_sr = {
198 .description = NULL_IF_CONFIG_SMALL("Apply DNN-based image super resolution to the input."),
199 .priv_size = sizeof(SRContext),
202 .query_formats = query_formats,
204 .outputs = sr_outputs,
205 .priv_class = &sr_class,