From patchwork Wed May 29 12:20:09 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Xuewei Meng X-Patchwork-Id: 13325 Return-Path: X-Original-To: patchwork@ffaux-bg.ffmpeg.org Delivered-To: patchwork@ffaux-bg.ffmpeg.org Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100]) by ffaux.localdomain (Postfix) with ESMTP id 5C174449A39 for ; Wed, 29 May 2019 15:20:23 +0300 (EEST) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 2BA7D68A65A; Wed, 29 May 2019 15:20:23 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mail-pl1-f170.google.com (mail-pl1-f170.google.com [209.85.214.170]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 71EC068A656 for ; Wed, 29 May 2019 15:20:16 +0300 (EEST) Received: by mail-pl1-f170.google.com with SMTP id g21so1009004plq.0 for ; Wed, 29 May 2019 05:20:16 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=from:to:subject:date:message-id:in-reply-to:references; bh=4xTuEueK31vVfcIKVunIVKyvNBneDBSBqgG776mRVyU=; b=jgvcsg3/cj2foOsgWgyCMbOZAZjwfdzHV/FssAvTuOv0XOQp3JpQdw+jGVhGBLOgJv P6eHF/dv5FxJ6JzPFSO7y7GC+gXbP5hDOw7Z2Hnd1MiQ3YKYUuVQ13Y7ky39zKdzh6dy 4MHEWXqeX1UWzU6i1Cir5Nen+6x5yL8UKNd5TjgE5Qi7Xxdm/lHknBmHZaKP2puoIXsw Sd/NS51pT9tZpnBAQs9QlrJHtFgvog4uRA++X3qlJrh3REWOxnm38rO+jEs6/Htz1Suc 5Qj3yOV4TKFlYTkLA7Vh1ZXhSgvayaHlEg5ejLREUe7dW6zejC8PBmiwLQI6ioIcmU84 Jl9g== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:subject:date:message-id:in-reply-to :references; bh=4xTuEueK31vVfcIKVunIVKyvNBneDBSBqgG776mRVyU=; b=m8cLzP17BoO5KxT+7FX01e91tQIK7JD7egnOvygrLOuzW6IGHyWcEKf8FWBB8IovIw YgnyvQn9KxHgBbL45YRcu9wnZPQ3UzElRBu6PrgHbnqkqXhPM8MhY7QNyQmjTmglDDLr TtljYcFeGNz4UcxcRIgH1RtrylDEM1gYQlZ5ihRYsvTKmxbqAHzfxX4SzfEhCoHSvmFc uBNMdcPVJHGNQjP7IO6NF+rUKFR6o2ma5xEPboK11N0tZbNsqCiBpOaR364E0rATcRr5 Tmyzi3/kpacPmZwqG4nd/rurqLBM6/LASwCrqXtsaPws4o1YhNCvDVxLWEJkWhibghzp Y/Sw== X-Gm-Message-State: APjAAAW/gEIdVOwV1Deazo/ZI6LtKynL6wsIJyqNxvMrejSJhBNTTmMe s5w4MYv+IKSVeEbPJtWc23GaD8xWAmo= X-Google-Smtp-Source: APXvYqxY9dsnOD2IAOaPx/vwozmAHv+N8hk6BZtwb+eH66pWMT0HgLQvS2DfRN4nLMetl1qeVRbWsg== X-Received: by 2002:a17:902:8209:: with SMTP id x9mr13671981pln.327.1559132414039; Wed, 29 May 2019 05:20:14 -0700 (PDT) Received: from DESKTOP-IACK8OK.localdomain ([2001:da8:201:3474:8091:8948:66b:5044]) by smtp.gmail.com with ESMTPSA id p13sm1432991pff.2.2019.05.29.05.20.12 for (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Wed, 29 May 2019 05:20:13 -0700 (PDT) From: Xuewei Meng To: ffmpeg-devel@ffmpeg.org Date: Wed, 29 May 2019 20:20:09 +0800 Message-Id: <20190529122009.83-1-xwmeng96@gmail.com> X-Mailer: git-send-email 2.17.1 In-Reply-To: References: Subject: [FFmpeg-devel] [PATCH v2] libavfilter: Add derain filter X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" Remove the rain in the input image/video by applying the derain methods based on convolutional neural networks. Training scripts as well as scripts for model generation are provided in the repository at https://github.com/XueweiMeng/derain_filter.git. Signed-off-by: Xuewei Meng --- doc/filters.texi | 34 ++++++ libavfilter/Makefile | 1 + libavfilter/allfilters.c | 1 + libavfilter/vf_derain.c | 216 +++++++++++++++++++++++++++++++++++++++ 4 files changed, 252 insertions(+) create mode 100644 libavfilter/vf_derain.c diff --git a/doc/filters.texi b/doc/filters.texi index 4fdcfe919e..f1d3841ed3 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -8248,6 +8248,40 @@ delogo=x=0:y=0:w=100:h=77:band=10 @end itemize +@section derain + +Remove the rain in the input image/video by applying the derain methods based on +convolutional neural networks. Supported models: + +@itemize +@item +Recurrent Squeeze-and-Excitation Context Aggregation Net (RESCAN). +See @url{http://openaccess.thecvf.com/content_ECCV_2018/papers/Xia_Li_Recurrent_Squeeze-and-Excitation_Context_ECCV_2018_paper.pdf}. +@end itemize + +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. +@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 9a61c25b05..b7191d0081 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -200,6 +200,7 @@ OBJS-$(CONFIG_DCTDNOIZ_FILTER) += vf_dctdnoiz.o OBJS-$(CONFIG_DEBAND_FILTER) += vf_deband.o OBJS-$(CONFIG_DEBLOCK_FILTER) += vf_deblock.o OBJS-$(CONFIG_DECIMATE_FILTER) += vf_decimate.o +OBJS-$(CONFIG_DERAIN_FILTER) += vf_derain.o OBJS-$(CONFIG_DECONVOLVE_FILTER) += vf_convolve.o framesync.o OBJS-$(CONFIG_DEDOT_FILTER) += vf_dedot.o OBJS-$(CONFIG_DEFLATE_FILTER) += vf_neighbor.o diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c index 40534738ee..f3c8883960 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -196,6 +196,7 @@ extern AVFilter ff_vf_deinterlace_vaapi; extern AVFilter ff_vf_dejudder; extern AVFilter ff_vf_delogo; extern AVFilter ff_vf_denoise_vaapi; +extern AVFilter ff_vf_derain; extern AVFilter ff_vf_deshake; extern AVFilter ff_vf_despill; extern AVFilter ff_vf_detelecine; diff --git a/libavfilter/vf_derain.c b/libavfilter/vf_derain.c new file mode 100644 index 0000000000..f7bbf314e5 --- /dev/null +++ b/libavfilter/vf_derain.c @@ -0,0 +1,216 @@ +/* + * 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. + * http://openaccess.thecvf.com/content_ECCV_2018/html/Xia_Li_Recurrent_Squeeze-and-Excitation_Context_ECCV_2018_paper.html + */ + +#include "libavformat/avio.h" +#include "libavutil/opt.h" +#include "avfilter.h" +#include "dnn_interface.h" +#include "formats.h" +#include "internal.h" + +typedef struct DRContext { + const AVClass *class; + + char *model_filename; + DNNBackendType backend_type; + DNNModule *dnn_module; + DNNModel *model; + DNNInputData input; + DNNData output; +} DRContext; + +#define CLIP(x, min, max) (x < min ? min : (x > max ? max : x)) +#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_INT, { .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; + const char *model_output_name = "y"; + 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, "x", &model_output_name, 1); + if (result != DNN_SUCCESS) { + av_log(ctx, AV_LOG_ERROR, "could not set input and output for the model\n"); + return AVERROR(EIO); + } + + 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; + int pad_size; + + 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); + + for (int i = 0; i < in->height; i++){ + for(int j = 0; j < in->width * 3; j++){ + int k = i * in->linesize[0] + j; + int t = i * in->width * 3 + j; + ((float *)dr_context->input.data)[t] = in->data[0][k] / 255.0; + } + } + + dnn_result = (dr_context->dnn_module->execute_model)(dr_context->model, &dr_context->output, 1); + if (dnn_result != DNN_SUCCESS){ + av_log(ctx, AV_LOG_ERROR, "failed to execute model\n"); + return AVERROR(EIO); + } + + out->height = dr_context->output.height; + out->width = dr_context->output.width; + outlink->h = dr_context->output.height; + outlink->w = dr_context->output.width; + pad_size = (in->height - out->height) >> 1; + + for (int i = 0; i < out->height; i++){ + for(int j = 0; j < out->width * 3; j++){ + int k = i * out->linesize[0] + j; + int t = i * out->width * 3 + j; + + int t_in = (i + pad_size) * in->width * 3 + j + pad_size * 3; + out->data[0][k] = CLIP((int)((((float *)dr_context->input.data)[t_in] - dr_context->output.data[t]) * 255), 0, 255); + } + } + + av_frame_free(&in); + + return ff_filter_frame(outlink, out); +} + +static av_cold int init(AVFilterContext *ctx) +{ + DRContext *dr_context = ctx->priv; + + dr_context->input.dt = DNN_FLOAT; + 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, +}; \ No newline at end of file