Hi, Yes, I use the espcn model for deraining as the initial version as it's a easier way to implement the filter, although the paper proposes it for super-resolution. And the model does have some effect on deraining project. While, it is just the first version. I will use more suitable and more powerful model for derain filter according to the latest models proposed in derain task, and I will upload the new model soon. As for the model training source code, I did develop the derain training code initially based on the sr model training code in order to confirm the feasibility of our method quickly. And sorry, I forgot to include the original author copyrights. I have been writing the model training code by myself, and will upload it soon. Thanks for your suggestion!
Xuewei > -----原始邮件----- > 发件人: "Pedro Arthur" <bygran...@gmail.com> > 发送时间: 2019-04-10 01:21:06 (星期三) > 收件人: "FFmpeg development discussions and patches" <ffmpeg-devel@ffmpeg.org> > 抄送: "Steven Liu" <l...@chinaffmpeg.org> > 主题: Re: [FFmpeg-devel] [PATCH] libavfilter: Add derain filter init > version--GSoC Qualification Task. > > Hi, > > Em ter, 9 de abr de 2019 às 04:15, <xwm...@pku.edu.cn> escreveu: > > +@section derain > > + > > +Remove the rain in the input image/video by applying the derain methods > > based on > > +convolutional neural networks. Supported models: > > + > > +@itemize > > +@item > > +Efficient Sub-Pixel Convolutional Neural Network model (ESPCN). > > +See @url{https://arxiv.org/abs/1609.05158}. > > +@end itemize > > As the doc suggests, you're using the espcn model for deraining? if > so, it would be more relevant to link to paper which justifies this > usage as it currently seems to suggest you're using super-resolution. > > In case you are the one which is proposing this usage, it worth at > least give some justification. is it better the current methods in any > way? > > > > + > > +Training scripts as well as scripts for model generation are provided in > > +the repository at @url{https://github.com/XueweiMeng/derain_filter.git}. > > + > > +The filter accepts the following options: > > + > > +@table @option > > +@item dnn_backend > > +Specify which DNN backend to use for model loading and execution. This > > option accepts > > +the following values: > > + > > +@table @samp > > +@item native > > +Native implementation of DNN loading and execution. > > + > > +@item tensorflow > > +TensorFlow backend. To enable this backend you > > +need to install the TensorFlow for C library (see > > +@url{https://www.tensorflow.org/install/install_c}) and configure FFmpeg > > with > > +@code{--enable-libtensorflow} > > +@end table > > + > > +Default value is @samp{native}. > > + > > +@item model > > +Set path to model file specifying network architecture and its parameters. > > +Note that different backends use different file formats. TensorFlow backend > > +can load files for both formats, while native backend can load files for > > only > > +its format. > > +@end table > > + > > @section deshake > > > > Attempt to fix small changes in horizontal and/or vertical shift. This > > diff --git a/libavfilter/Makefile b/libavfilter/Makefile > > index fef6ec5c55..7809bac565 100644 > > --- a/libavfilter/Makefile > > +++ b/libavfilter/Makefile > > @@ -194,6 +194,7 @@ OBJS-$(CONFIG_DATASCOPE_FILTER) += > > vf_datascope.o > > OBJS-$(CONFIG_DCTDNOIZ_FILTER) += vf_dctdnoiz.o > > OBJS-$(CONFIG_DEBAND_FILTER) += vf_deband.o > > OBJS-$(CONFIG_DEBLOCK_FILTER) += vf_deblock.o > > +OBJS-$(CONFIG_DERAIN_FILTER) += vf_derain.o > > OBJS-$(CONFIG_DECIMATE_FILTER) += vf_decimate.o > > OBJS-$(CONFIG_DECONVOLVE_FILTER) += vf_convolve.o framesync.o > > OBJS-$(CONFIG_DEDOT_FILTER) += vf_dedot.o > > diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c > > index c51ae0f3c7..ee2a5b63e6 100644 > > --- a/libavfilter/allfilters.c > > +++ b/libavfilter/allfilters.c > > @@ -182,6 +182,7 @@ extern AVFilter ff_vf_datascope; > > extern AVFilter ff_vf_dctdnoiz; > > extern AVFilter ff_vf_deband; > > extern AVFilter ff_vf_deblock; > > +extern AVFilter ff_vf_derain; > > extern AVFilter ff_vf_decimate; > > extern AVFilter ff_vf_deconvolve; > > extern AVFilter ff_vf_dedot; > > diff --git a/libavfilter/vf_derain.c b/libavfilter/vf_derain.c > > new file mode 100644 > > index 0000000000..f72ae1cd3a > > --- /dev/null > > +++ b/libavfilter/vf_derain.c > > @@ -0,0 +1,204 @@ > > +/* > > + * Copyright (c) 2019 Xuewei Meng > > + * > > + * 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 image derain filter using deep convolutional > > networks. > > + * https://arxiv.org/abs/1609.05158 > > + * > > http://openaccess.thecvf.com/content_ECCV_2018/html/Xia_Li_Recurrent_Squeeze-and-Excitation_Context_ECCV_2018_paper.html > > + */ > > + > > +#include "libavutil/opt.h" > > +#include "libavformat/avio.h" > > +#include "libswscale/swscale.h" > > +#include "avfilter.h" > > +#include "formats.h" > > +#include "internal.h" > > +#include "dnn_interface.h" > > + > > +typedef struct DRContext { > > + const AVClass *class; > > + > > + char *model_filename; > > + DNNBackendType backend_type; > > + DNNModule *dnn_module; > > + DNNModel *model; > > + DNNData input; > > + DNNData output; > > +} DRContext; > > + > > +#define OFFSET(x) offsetof(DRContext, x) > > +#define FLAGS AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM > > +static const AVOption derain_options[] = { > > + { "dnn_backend", "DNN backend", OFFSET(backend_type), > > AV_OPT_TYPE_FLAGS, { .i64 = 0 }, 0, 1, FLAGS, "backend" }, > > + { "native", "native backend flag", 0, > > AV_OPT_TYPE_CONST, { .i64 = 0 }, 0, 0, FLAGS, "backend" }, > > +#if (CONFIG_LIBTENSORFLOW == 1) > > + { "tensorflow", "tensorflow backend flag", 0, > > AV_OPT_TYPE_CONST, { .i64 = 1 }, 0, 0, FLAGS, "backend" }, > > +#endif > > + { "model", "path to model file", OFFSET(model_filename), > > AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, FLAGS }, > > + { NULL } > > +}; > > + > > +AVFILTER_DEFINE_CLASS(derain); > > + > > +static int query_formats(AVFilterContext *ctx) > > +{ > > + AVFilterFormats *formats; > > + const enum AVPixelFormat pixel_fmts[] = { > > + AV_PIX_FMT_RGB24, > > + AV_PIX_FMT_NONE > > + }; > > + > > + formats = ff_make_format_list(pixel_fmts); > > + if (!formats) { > > + av_log(ctx, AV_LOG_ERROR, "could not create formats list\n"); > > + return AVERROR(ENOMEM); > > + } > > + > > + return ff_set_common_formats(ctx, formats); > > +} > > + > > +static int config_inputs(AVFilterLink *inlink) > > +{ > > + AVFilterContext *ctx = inlink->dst; > > + DRContext *dr_context = ctx->priv; > > + AVFilterLink *outlink = ctx->outputs[0]; > > + DNNReturnType result; > > + > > + dr_context->input.width = inlink->w; > > + dr_context->input.height = inlink->h; > > + dr_context->input.channels = 3; > > + > > + result = > > (dr_context->model->set_input_output)(dr_context->model->model, > > &dr_context->input, &dr_context->output); > > + if (result != DNN_SUCCESS) { > > + av_log(ctx, AV_LOG_ERROR, "could not set input and output for the > > model\n"); > > + return AVERROR(EIO); > > + } > > + > > + outlink->h = dr_context->output.height; > > + outlink->w = dr_context->output.width; > > + > > + return 0; > > +} > > + > > +static int filter_frame(AVFilterLink *inlink, AVFrame *in) > > +{ > > + AVFilterContext *ctx = inlink->dst; > > + AVFilterLink *outlink = ctx->outputs[0]; > > + DRContext *dr_context = ctx->priv; > > + DNNReturnType dnn_result; > > + > > + AVFrame *out = ff_get_video_buffer(outlink, outlink->w, outlink->h); > > + if (!out) { > > + av_log(ctx, AV_LOG_ERROR, "could not allocate memory for output > > frame\n"); > > + av_frame_free(&in); > > + return AVERROR(ENOMEM); > > + } > > + > > + av_frame_copy_props(out, in); > > + out->height = dr_context->output.height; > > + out->width = dr_context->output.width; > > + > > + for (int i = 0; i < out->height * out->width * 3; i++) { > > + dr_context->input.data[i] = in->data[0][i] / 255.0; > > + } > > + > > + av_frame_free(&in); > > + dnn_result = > > (dr_context->dnn_module->execute_model)(dr_context->model); > > + if (dnn_result != DNN_SUCCESS){ > > + av_log(ctx, AV_LOG_ERROR, "failed to execute model\n"); > > + return AVERROR(EIO); > > + } > > + > > + for (int i = 0; i < out->height * out->width * 3; i++) { > > + out->data[0][i] = (int)(dr_context->output.data[i] * 255); > > + } > > + > > + return ff_filter_frame(outlink, out); > > +} > > + > > +static av_cold int init(AVFilterContext *ctx) > > +{ > > + DRContext *dr_context = ctx->priv; > > + > > + dr_context->dnn_module = ff_get_dnn_module(dr_context->backend_type); > > + if (!dr_context->dnn_module) { > > + av_log(ctx, AV_LOG_ERROR, "could not create DNN module for > > requested backend\n"); > > + return AVERROR(ENOMEM); > > + } > > + if (!dr_context->model_filename) { > > + av_log(ctx, AV_LOG_ERROR, "model file for network is not > > specified\n"); > > + return AVERROR(EINVAL); > > + } > > + if (!dr_context->dnn_module->load_model) { > > + av_log(ctx, AV_LOG_ERROR, "load_model for network is not > > specified\n"); > > + return AVERROR(EINVAL); > > + } > > + > > + dr_context->model = > > (dr_context->dnn_module->load_model)(dr_context->model_filename); > > + if (!dr_context->model) { > > + av_log(ctx, AV_LOG_ERROR, "could not load DNN model\n"); > > + return AVERROR(EINVAL); > > + } > > + > > + return 0; > > +} > > + > > +static av_cold void uninit(AVFilterContext *ctx) > > +{ > > + DRContext *dr_context = ctx->priv; > > + > > + if (dr_context->dnn_module) { > > + (dr_context->dnn_module->free_model)(&dr_context->model); > > + av_freep(&dr_context->dnn_module); > > + } > > +} > > + > > +static const AVFilterPad derain_inputs[] = { > > + { > > + .name = "default", > > + .type = AVMEDIA_TYPE_VIDEO, > > + .config_props = config_inputs, > > + .filter_frame = filter_frame, > > + }, > > + { NULL } > > +}; > > + > > +static const AVFilterPad derain_outputs[] = { > > + { > > + .name = "default", > > + .type = AVMEDIA_TYPE_VIDEO, > > + }, > > + { NULL } > > +}; > > + > > +AVFilter ff_vf_derain = { > > + .name = "derain", > > + .description = NULL_IF_CONFIG_SMALL("Apply derain filter to the > > input."), > > + .priv_size = sizeof(DRContext), > > + .init = init, > > + .uninit = uninit, > > + .query_formats = query_formats, > > + .inputs = derain_inputs, > > + .outputs = derain_outputs, > > + .priv_class = &derain_class, > > + .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | > > AVFILTER_FLAG_SLICE_THREADS, > > +}; > > + > > -- > > 2.17.1 > > > > _______________________________________________ > > ffmpeg-devel mailing list > > ffmpeg-devel@ffmpeg.org > > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > > > To unsubscribe, visit link above, or email > > ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe". > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe". _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".