2018-05-07 0:30 GMT-03:00 Steven Liu <l...@chinaffmpeg.org>: > Hi Sergey, > > How should i test this filter? > I tested it some days ago, the picture get worse from 2nd frame. > input resolution 640x480 to 1280x720; > > ffmpeg -i input -vf srcnn output Hi, The filter expects the input upscaled by 2x, therefore the proper command would be
ffmpeg -i input -vf "scale=2*iw:2*ih, srcnn, scale=1280:720" The default filter is trained for 2x upscale, anything different from that may generate bad results. > >> On May 7, 2018, at 09:33, Pedro Arthur <bygran...@gmail.com> wrote: >> >> 2018-05-06 17:27 GMT-03:00 James Almer <jamr...@gmail.com>: >>> On 5/5/2018 5:38 PM, James Almer wrote: >>>> On 4/10/2018 2:16 PM, Sergey Lavrushkin wrote: >>>>> diff --git a/libavfilter/vf_srcnn.c b/libavfilter/vf_srcnn.c >>>>> new file mode 100644 >>>>> index 0000000000..d9b4891f7f >>>>> --- /dev/null >>>>> +++ b/libavfilter/vf_srcnn.c >>>>> @@ -0,0 +1,420 @@ >>>>> +/* >>>>> + * Copyright (c) 2018 Sergey Lavrushkin >>>>> + * >>>>> + * 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 super-resolution using deep convolutional >>>>> networks. >>>>> + * https://arxiv.org/abs/1501.00092 >>>>> + */ >>>>> + >>>>> +#include "avfilter.h" >>>>> +#include "formats.h" >>>>> +#include "internal.h" >>>>> +#include "libavutil/opt.h" >>>>> +#include "unistd.h" >>>> >>>> This broke msvc. It should be (if anything) <unistd.h>, and wrapped >>>> inside a HAVE_UNISTD_H preprocessor check. >>> >>> I (accidentally) applied this change as part of an unrelated commit, so >>> that's dealt with at least. >>> >>>> >>>>> +static av_cold int init(AVFilterContext* context) >>>>> +{ >>>>> + SRCNNContext *srcnn_context = context->priv; >>>>> + AVIOContext* config_file_context; >>>>> + int64_t file_size, srcnn_size; >>>>> + >>>>> + /// Check specified confguration file name and read network weights >>>>> from it >>>>> + if (!srcnn_context->config_file_path){ >>>>> + av_log(context, AV_LOG_INFO, "configuration file for network was >>>>> not specified, using default weights for x2 upsampling\n"); >>>>> + >>>>> + /// Create convolution kernels and copy default weights >>>>> + srcnn_context->conv1.input_channels = 1; >>>>> + srcnn_context->conv1.output_channels = 64; >>>>> + srcnn_context->conv1.size = 9; >>>>> + CHECK_ALLOCATION(allocate_copy_conv_data(&srcnn_context->conv1, >>>>> conv1_kernel, conv1_biases), ) >>>>> + >>>>> + srcnn_context->conv2.input_channels = 64; >>>>> + srcnn_context->conv2.output_channels = 32; >>>>> + srcnn_context->conv2.size = 1; >>>>> + CHECK_ALLOCATION(allocate_copy_conv_data(&srcnn_context->conv2, >>>>> conv2_kernel, conv2_biases), ) >>>>> + >>>>> + srcnn_context->conv3.input_channels = 32; >>>>> + srcnn_context->conv3.output_channels = 1; >>>>> + srcnn_context->conv3.size = 5; >>>>> + CHECK_ALLOCATION(allocate_copy_conv_data(&srcnn_context->conv3, >>>>> conv3_kernel, conv3_biases), ) >>>>> + } >>>>> + else if (access(srcnn_context->config_file_path, R_OK) != -1){ >>>> >>>> This seems to be the function needed from unistd.h >>>> >>>> Looking at configure and libavformat/file.c it's apparently not >>>> available everywhere (we have a HAVE_ACCESS define, and also check for >>>> mode constants like R_OK). >>> >>> MSVC does not have access(), so i have temporarily made this filter >>> depend on its presence until a proper fix is applied. That target has >>> been broken for a few days now and that shouldn't be the case. >>> It should be a matter or doing the same thing as in libavformat/file.c, >>> i guess, or perhaps use _access() instead, which is available in MSVC. >> Thanks. >> >> >> @Sergey can you provide a definitive fix? >> >>> _______________________________________________ >>> ffmpeg-devel mailing list >>> ffmpeg-devel@ffmpeg.org >>> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel >> _______________________________________________ >> ffmpeg-devel mailing list >> ffmpeg-devel@ffmpeg.org >> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > Thanks > Steven > > > > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
