[FFmpeg-devel] [PATCH v2] libavfilter: Add derain filter
Xuewei Meng
xwmeng96 at gmail.com
Wed May 29 15:20:09 EEST 2019
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 <xwmeng96 at gmail.com>
---
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
+ at section derain
+
+Remove the rain in the input image/video by applying the derain methods based on
+convolutional neural networks. Supported models:
+
+ at itemize
+ at 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}.
+ at 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:
+
+ at table @option
+ at item dnn_backend
+Specify which DNN backend to use for model loading and execution. This option accepts
+the following values:
+
+ at table @samp
+ at item native
+Native implementation of DNN loading and execution.
+ at end table
+Default value is @samp{native}.
+
+ at 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.
+ at 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
--
2.17.1
More information about the ffmpeg-devel
mailing list