[FFmpeg-devel] [PATCH V4] Add a filter implementing HDR image generation from a single exposure using deep CNNs
Guo, Yejun
yejun.guo at intel.com
Fri Nov 9 15:07:47 EET 2018
this filter accepts 8bit frame (RGB24) and outputs 10bit/float frame, and there's no reference image, so it is not feasible to use criteria such as PNSR, SSIM.
I choose the same method described in the paper to demo the filter effect, that means the frames before/after the filter are reduced by 3 stops.
The native video (test.native.mp4) is created from 7 png files @ https://github.com/gabrieleilertsen/hdrcnn/tree/master/data (the size of the image is enlarged to 1920*1080 with extra area filled with white) with command line: ffmpeg -f image2 -i ./img_%03d.png -c:v libx264 -preset veryslow -crf 1 test.native.mp4.
And two rgb24 videos are generated before/after the filter with -3 stops by modifying the code a little, see in the video folder at https://drive.google.com/drive/folders/1URsRY5g-VdE-kHlP5vQoLoimMIZ-SX00?usp=sharing
for your convenient, I also dump png files from generated videos and combine the before/after pngs into one file, see in png folder at the google drive.
> -----Original Message-----
> From: ffmpeg-devel [mailto:ffmpeg-devel-bounces at ffmpeg.org] On Behalf
> Of Liu Steven
> Sent: Monday, November 05, 2018 3:57 PM
> To: FFmpeg development discussions and patches <ffmpeg-
> devel at ffmpeg.org>
> Cc: Liu Steven <lq at chinaffmpeg.org>
> Subject: Re: [FFmpeg-devel] [PATCH V4] Add a filter implementing HDR
> image generation from a single exposure using deep CNNs
>
>
>
> > 在 2018年11月5日，下午3:42，Guo, Yejun <yejun.guo at intel.com> 写
> 道：
> >
> > ask for comment or merge, thanks.
> Will push after 24 hours if there have no objections.
> >
> >> -----Original Message-----
> >> From: ffmpeg-devel [mailto:ffmpeg-devel-bounces at ffmpeg.org] On
> Behalf
> >> Of Guo, Yejun
> >> Sent: Monday, October 29, 2018 11:19 AM
> >> To: ffmpeg-devel at ffmpeg.org
> >> Subject: Re: [FFmpeg-devel] [PATCH V4] Add a filter implementing HDR
> >> image generation from a single exposure using deep CNNs
> >>
> >> any more comment? thanks.
> >>
> >>> -----Original Message-----
> >>> From: Guo, Yejun
> >>> Sent: Tuesday, October 23, 2018 6:46 AM
> >>> To: ffmpeg-devel at ffmpeg.org
> >>> Cc: Guo, Yejun <yejun.guo at intel.com>; Guo
> >>> Subject: [PATCH V4] Add a filter implementing HDR image generation
> >>> from a single exposure using deep CNNs
> >>>
> >>> see the algorithm's paper and code below.
> >>>
> >>> the filter's parameter looks like:
> >>>
> >>
> sdr2hdr=model_filename=/path_to_tensorflow_graph.pb:out_fmt=gbrp10l
> >>> e
> >>>
> >>> The input of the deep CNN model is RGB24 while the output is float
> >>> for each color channel. This is the filter's default behavior to
> >>> output format with gbrpf32le. And gbrp10le is also supported as the
> >>> output, so we can see the rendering result in a player, as a reference.
> >>>
> >>> To generate the model file, we need modify the original script a little.
> >>> - set name='y' for y_final within script at
> >>> https://github.com/gabrieleilertsen/hdrcnn/blob/master/network.py
> >>> - add the following code to the script at
> >>>
> https://github.com/gabrieleilertsen/hdrcnn/blob/master/hdrcnn_predict.
> >>> py
> >>>
> >>> graph = tf.graph_util.convert_variables_to_constants(sess,
> >>> sess.graph_def,
> >>> ["y"]) tf.train.write_graph(graph, '.', 'graph.pb', as_text=False)
> >>>
> >>> The filter only works when tensorflow C api is supported in the
> >>> system, native backend is not supported since there are some
> >>> different types of layers in the deep CNN model, besides CONV and
> >> DEPTH_TO_SPACE.
> >>>
> >>> https://arxiv.org/pdf/1710.07480.pdf:
> >>> author = "Eilertsen, Gabriel and Kronander, Joel, and Denes, Gyorgy
> >> and
> >>> Mantiuk, Rafał and Unger, Jonas",
> >>> title = "HDR image reconstruction from a single exposure using deep
> >>> CNNs",
> >>> journal = "ACM Transactions on Graphics (TOG)",
> >>> number = "6",
> >>> volume = "36",
> >>> articleno = "178",
> >>> year = "2017"
> >>>
> >>> https://github.com/gabrieleilertsen/hdrcnn
> >>>
> >>> btw, as a whole solution, metadata should also be generated from the
> >>> sdr video, so to be encoded as a HDR video. Not supported yet.
> >>> This patch just focuses on this paper.
> >>>
> >>> Signed-off-by: Guo, Yejun <yejun.guo at intel.com>
> >>> ---
> >>> configure | 1 +
> >>> doc/filters.texi | 35 +++++++
> >>> libavfilter/Makefile | 1 +
> >>> libavfilter/allfilters.c | 1 +
> >>> libavfilter/vf_sdr2hdr.c | 268
> >>> +++++++++++++++++++++++++++++++++++++++++++++++
> >>> 5 files changed, 306 insertions(+)
> >>> create mode 100644 libavfilter/vf_sdr2hdr.c
> >>>
> >>> diff --git a/configure b/configure
> >>> index 85d5dd5..5e2efba 100755
> >>> --- a/configure
> >>> +++ b/configure
> >>> @@ -3438,6 +3438,7 @@ scale2ref_filter_deps="swscale"
> >>> scale_filter_deps="swscale"
> >>> scale_qsv_filter_deps="libmfx"
> >>> select_filter_select="pixelutils"
> >>> +sdr2hdr_filter_deps="libtensorflow"
> >>> sharpness_vaapi_filter_deps="vaapi"
> >>> showcqt_filter_deps="avcodec avformat swscale"
> >>> showcqt_filter_suggest="libfontconfig libfreetype"
> >>> diff --git a/doc/filters.texi b/doc/filters.texi index
> >>> 17e2549..bba9f87 100644
> >>> --- a/doc/filters.texi
> >>> +++ b/doc/filters.texi
> >>> @@ -14672,6 +14672,41 @@ Scale a subtitle stream (b) to match the
> >>> main video (a) in size before overlayin @end example @end itemize
> >>>
> >>> + at section sdr2hdr
> >>> +
> >>> +HDR image generation from a single exposure using deep CNNs with
> >>> TensorFlow C library.
> >>> +
> >>> + at itemize
> >>> + at item
> >>> +paper: see @url{https://arxiv.org/pdf/1710.07480.pdf}
> >>> +
> >>> + at item
> >>> +code with model and trained parameters: see
> >>> + at url{https://github.com/gabrieleilertsen/hdrcnn}
> >>> + at end itemize
> >>> +
> >>> +The filter accepts the following options:
> >>> +
> >>> + at table @option
> >>> +
> >>> + at item model_filename
> >>> +Set path to model file specifying network architecture and its
> parameters.
> >>> +
> >>> + at item out_fmt
> >>> +the data format of the filter's output.
> >>> +
> >>> +It accepts the following values:
> >>> + at table @samp
> >>> + at item gbrpf32le
> >>> +force gbrpf32le output
> >>> +
> >>> + at item gbrp10le
> >>> +force gbrp10le output
> >>> + at end table
> >>> +
> >>> +Default value is @samp{gbrpf32le}.
> >>> +
> >>> + at end table
> >>> +
> >>> @anchor{selectivecolor}
> >>> @section selectivecolor
> >>>
> >>> diff --git a/libavfilter/Makefile b/libavfilter/Makefile index
> >>> 62cc2f5..88e7da6
> >>> 100644
> >>> --- a/libavfilter/Makefile
> >>> +++ b/libavfilter/Makefile
> >>> @@ -360,6 +360,7 @@ OBJS-$(CONFIG_SOBEL_OPENCL_FILTER) +=
> >>> vf_convolution_opencl.o opencl.o
> >>> OBJS-$(CONFIG_SPLIT_FILTER) += split.o
> >>> OBJS-$(CONFIG_SPP_FILTER) += vf_spp.o
> >>> OBJS-$(CONFIG_SR_FILTER) += vf_sr.o
> >>> +OBJS-$(CONFIG_SDR2HDR_FILTER) += vf_sdr2hdr.o
> >>> OBJS-$(CONFIG_SSIM_FILTER) += vf_ssim.o framesync.o
> >>> OBJS-$(CONFIG_STEREO3D_FILTER) += vf_stereo3d.o
> >>> OBJS-$(CONFIG_STREAMSELECT_FILTER) += f_streamselect.o
> >>> framesync.o
> >>> diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
> >>> index 5e72803..1645c0f
> >>> 100644
> >>> --- a/libavfilter/allfilters.c
> >>> +++ b/libavfilter/allfilters.c
> >>> @@ -319,6 +319,7 @@ extern AVFilter ff_vf_scale_npp; extern
> >>> AVFilter ff_vf_scale_qsv; extern AVFilter ff_vf_scale_vaapi;
> >>> extern AVFilter ff_vf_scale2ref;
> >>> +extern AVFilter ff_vf_sdr2hdr;
> >>> extern AVFilter ff_vf_select;
> >>> extern AVFilter ff_vf_selectivecolor; extern AVFilter
> >>> ff_vf_sendcmd; diff --git a/libavfilter/vf_sdr2hdr.c
> >>> b/libavfilter/vf_sdr2hdr.c new file mode
> >>> 100644 index 0000000..109b907
> >>> --- /dev/null
> >>> +++ b/libavfilter/vf_sdr2hdr.c
> >>> @@ -0,0 +1,268 @@
> >>> +/*
> >>> + * Copyright (c) 2018 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
> >>> + * Filter implementing HDR image generation from a single exposure
> >>> +using
> >>> deep CNNs.
> >>> + * https://arxiv.org/pdf/1710.07480.pdf
> >>> + */
> >>> +
> >>> +#include "avfilter.h"
> >>> +#include "formats.h"
> >>> +#include "internal.h"
> >>> +#include "libavutil/opt.h"
> >>> +#include "libavutil/qsort.h"
> >>> +#include "libavformat/avio.h"
> >>> +#include "libswscale/swscale.h"
> >>> +#include "dnn_interface.h"
> >>> +#include <math.h>
> >>> +
> >>> +typedef struct SDR2HDRContext {
> >>> + const AVClass *class;
> >>> +
> >>> + char* model_filename;
> >>> + enum AVPixelFormat out_fmt;
> >>> + DNNModule* dnn_module;
> >>> + DNNModel* model;
> >>> + DNNData input, output;
> >>> +} SDR2HDRContext;
> >>> +
> >>> +#define OFFSET(x) offsetof(SDR2HDRContext, x) #define FLAGS
> >>> +AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM
> >> static
> >>> const
> >>> +AVOption sdr2hdr_options[] = {
> >>> + { "model_filename", "path to model file specifying network
> >>> +architecture
> >>> and its parameters", OFFSET(model_filename), AV_OPT_TYPE_STRING,
> >>> {.str=NULL}, 0, 0, FLAGS },
> >>> + { "out_fmt", "the data format of the filter's output, it could
> >>> + be gbrpf32le
> >>> [default] or gbrp10le", OFFSET(out_fmt), AV_OPT_TYPE_PIXEL_FMT,
> >>> {.i64=AV_PIX_FMT_GBRPF32LE}, AV_PIX_FMT_NONE, AV_PIX_FMT_NB,
> >> FLAGS },
> >>> + { NULL }
> >>> +};
> >>> +
> >>> +AVFILTER_DEFINE_CLASS(sdr2hdr);
> >>> +
> >>> +static av_cold int init(AVFilterContext* context) {
> >>> + SDR2HDRContext* ctx = context->priv;
> >>> +
> >>> + if (ctx->out_fmt != AV_PIX_FMT_GBRPF32LE && ctx->out_fmt !=
> >>> AV_PIX_FMT_GBRP10LE) {
> >>> + av_log(context, AV_LOG_ERROR, "could not support the output
> >>> format\n");
> >>> + return AVERROR(ENOSYS);
> >>> + }
> >>> +
> >>> + ctx->dnn_module = ff_get_dnn_module(DNN_TF);
> >>> + if (!ctx->dnn_module){
> >>> + av_log(context, AV_LOG_ERROR, "could not create DNN module
> >>> + for
> >>> tensorflow backend\n");
> >>> + return AVERROR(ENOMEM);
> >>> + }
> >>> + if (!ctx->model_filename){
> >>> + av_log(context, AV_LOG_ERROR, "model file for network was
> >>> + not
> >>> specified\n");
> >>> + return AVERROR(EIO);
> >>> + }
> >>> + if (!ctx->dnn_module->load_model) {
> >>> + av_log(context, AV_LOG_ERROR, "load_model for network was
> >>> + not
> >>> specified\n");
> >>> + return AVERROR(EIO);
> >>> + }
> >>> + ctx->model = (ctx->dnn_module->load_model)(ctx-
> >model_filename);
> >>> + if (!ctx->model){
> >>> + av_log(context, AV_LOG_ERROR, "could not load DNN model\n");
> >>> + return AVERROR(EIO);
> >>> + }
> >>> + return 0;
> >>> +}
> >>> +
> >>> +static int query_formats(AVFilterContext* context) {
> >>> + const enum AVPixelFormat in_formats[] = {AV_PIX_FMT_RGB24,
> >>> + AV_PIX_FMT_NONE};
> >>> + enum AVPixelFormat out_formats[2];
> >>> + SDR2HDRContext* ctx = context->priv;
> >>> + AVFilterFormats* formats_list;
> >>> + int ret = 0;
> >>> +
> >>> + formats_list = ff_make_format_list(in_formats);
> >>> + if ((ret = ff_formats_ref(formats_list,
> >>> + &context->inputs[0]->out_formats))
> >>> < 0)
> >>> + return ret;
> >>> +
> >>> + out_formats[0] = ctx->out_fmt;
> >>> + out_formats[1] = AV_PIX_FMT_NONE;
> >>> + formats_list = ff_make_format_list(out_formats);
> >>> + if ((ret = ff_formats_ref(formats_list,
> >>> + &context->outputs[0]->in_formats))
> >>> < 0)
> >>> + return ret;
> >>> +
> >>> + return 0;
> >>> +}
> >>> +
> >>> +static int config_props(AVFilterLink* inlink) {
> >>> + AVFilterContext* context = inlink->dst;
> >>> + SDR2HDRContext* ctx = context->priv;
> >>> + AVFilterLink* outlink = context->outputs[0];
> >>> + DNNReturnType result;
> >>> +
> >>> + // the dnn model is tied with resolution due to deconv layer of
> >> tensorflow
> >>> + // now just support 1920*1080 and so the magic numbers within this
> file
> >>> + if (inlink->w != 1920 || inlink->h != 1080) {
> >>> + av_log(context, AV_LOG_ERROR, "only support frame size with
> >>> 1920*1080\n");
> >>> + return AVERROR(ENOSYS);
> >>> + }
> >>> +
> >>> + ctx->input.width = 1920;
> >>> + ctx->input.height = 1088; //the model requires height is a multiple of
> 32,
> >>> + ctx->input.channels = 3;
> >>> +
> >>> + result = (ctx->model->set_input_output)(ctx->model->model,
> >>> + &ctx-
> >>>> input, &ctx->output);
> >>> + if (result != DNN_SUCCESS){
> >>> + av_log(context, AV_LOG_ERROR, "could not set input and
> >>> + output for
> >>> the model\n");
> >>> + return AVERROR(EIO);
> >>> + }
> >>> +
> >>> + memset(ctx->input.data, 0, ctx->input.channels *
> >>> + ctx->input.width
> >>> + * ctx-
> >>>> input.height * sizeof(float));
> >>> + outlink->h = 1080;
> >>> + outlink->w = 1920;
> >>> + return 0;
> >>> +}
> >>> +
> >>> +static float qsort_comparison_function_float(const void *a, const
> >>> +void
> >>> +*b) {
> >>> + return *(const float *)a - *(const float *)b; }
> >>> +
> >>> +static int filter_frame(AVFilterLink* inlink, AVFrame* in) {
> >>> + DNNReturnType dnn_result = DNN_SUCCESS;
> >>> + AVFilterContext* context = inlink->dst;
> >>> + SDR2HDRContext* ctx = context->priv;
> >>> + AVFilterLink* outlink = context->outputs[0];
> >>> + AVFrame* out = ff_get_video_buffer(outlink, outlink->w, outlink-
> >h);
> >>> + int total_pixels = in->height * in->width;
> >>> +
> >>> + if (!out){
> >>> + av_log(context, AV_LOG_ERROR, "could not allocate memory
> >>> + for
> >>> output frame\n");
> >>> + av_frame_free(&in);
> >>> + return AVERROR(ENOMEM);
> >>> + }
> >>> +
> >>> + av_frame_copy_props(out, in);
> >>> +
> >>> + for (int i = 0; i < in->linesize[0] * in->height; ++i) {
> >>> + ctx->input.data[i] = in->data[0][i] / 255.0f;
> >>> + }
> >>> +
> >>> + dnn_result = (ctx->dnn_module->execute_model)(ctx->model);
> >>> + if (dnn_result != DNN_SUCCESS){
> >>> + av_log(context, AV_LOG_ERROR, "failed to execute loaded
> >> model\n");
> >>> + return AVERROR(EIO);
> >>> + }
> >>> +
> >>> + if (ctx->out_fmt == AV_PIX_FMT_GBRPF32LE) {
> >>> + float* outg = (float*)out->data[0];
> >>> + float* outb = (float*)out->data[1];
> >>> + float* outr = (float*)out->data[2];
> >>> + for (int i = 0; i < total_pixels; ++i) {
> >>> + float r = ctx->output.data[i*3];
> >>> + float g = ctx->output.data[i*3+1];
> >>> + float b = ctx->output.data[i*3+2];
> >>> + outr[i] = r;
> >>> + outg[i] = g;
> >>> + outb[i] = b;
> >>> + }
> >>> + } else {
> >>> + // here, we just use a rough mapping to the 10bit contents
> >>> + // meta data generation for HDR video encoding is not supported
> yet
> >>> + float* converted_data = (float*)av_malloc(total_pixels * 3
> >>> + *
> >>> sizeof(float));
> >>> + int16_t* outg = (int16_t*)out->data[0];
> >>> + int16_t* outb = (int16_t*)out->data[1];
> >>> + int16_t* outr = (int16_t*)out->data[2];
> >>> +
> >>> + float max = 1.0f;
> >>> + for (int i = 0; i < total_pixels * 3; ++i) {
> >>> + float d = ctx->output.data[i];
> >>> + d = sqrt(d);
> >>> + converted_data[i] = d;
> >>> + max = FFMAX(d, max);
> >>> + }
> >>> +
> >>> + if (max > 1.0f) {
> >>> + AV_QSORT(converted_data, total_pixels * 3, float,
> >>> qsort_comparison_function_float);
> >>> + // 0.5% pixels are clipped
> >>> + max = converted_data[(int)(total_pixels * 3 * 0.995)];
> >>> + max = FFMAX(max, 1.0f);
> >>> +
> >>> + for (int i = 0; i < total_pixels * 3; ++i) {
> >>> + float d = ctx->output.data[i];
> >>> + d = sqrt(d);
> >>> + d = FFMIN(d, max);
> >>> + converted_data[i] = d;
> >>> + }
> >>> + }
> >>> +
> >>> + for (int i = 0; i < total_pixels; ++i) {
> >>> + float r = converted_data[i*3];
> >>> + float g = converted_data[i*3+1];
> >>> + float b = converted_data[i*3+2];
> >>> + outr[i] = r / max * 1023;
> >>> + outg[i] = g / max * 1023;
> >>> + outb[i] = b / max * 1023;
> >>> + }
> >>> +
> >>> + av_free(converted_data);
> >>> + }
> >>> +
> >>> + av_frame_free(&in);
> >>> + return ff_filter_frame(outlink, out); }
> >>> +
> >>> +static av_cold void uninit(AVFilterContext* context) {
> >>> + SDR2HDRContext* ctx = context->priv;
> >>> +
> >>> + if (ctx->dnn_module){
> >>> + (ctx->dnn_module->free_model)(&ctx->model);
> >>> + av_freep(&ctx->dnn_module);
> >>> + }
> >>> +}
> >>> +
> >>> +static const AVFilterPad sdr2hdr_inputs[] = {
> >>> + {
> >>> + .name = "default",
> >>> + .type = AVMEDIA_TYPE_VIDEO,
> >>> + .config_props = config_props,
> >>> + .filter_frame = filter_frame,
> >>> + },
> >>> + { NULL }
> >>> +};
> >>> +
> >>> +static const AVFilterPad sdr2hdr_outputs[] = {
> >>> + {
> >>> + .name = "default",
> >>> + .type = AVMEDIA_TYPE_VIDEO,
> >>> + },
> >>> + { NULL }
> >>> +};
> >>> +
> >>> +AVFilter ff_vf_sdr2hdr = {
> >>> + .name = "sdr2hdr",
> >>> + .description = NULL_IF_CONFIG_SMALL("HDR image generation
> from a
> >>> single exposure using deep CNNs."),
> >>> + .priv_size = sizeof(SDR2HDRContext),
> >>> + .init = init,
> >>> + .uninit = uninit,
> >>> + .query_formats = query_formats,
> >>> + .inputs = sdr2hdr_inputs,
> >>> + .outputs = sdr2hdr_outputs,
> >>> + .priv_class = &sdr2hdr_class,
> >>> + .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,
> >>> +};
> >>> --
> >>> 2.7.4
> >>
> >> _______________________________________________
> >> ffmpeg-devel mailing list
> >> ffmpeg-devel at ffmpeg.org
> >> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> > _______________________________________________
> > ffmpeg-devel mailing list
> > ffmpeg-devel at ffmpeg.org
> > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
More information about the ffmpeg-devel
mailing list