Wenbin,
Thank you for the advice. I spent three days trying to get it working. No
joy. I think I may have to move on to something else.
Enjoy the weekend!
Regards,
-Ron
-----Original Message-----
From: ffmpeg-user <ffmpeg-user-boun...@ffmpeg.org> On Behalf Of Chen, Wenbin
Sent: Thursday, April 18, 2024 10:56 PM
To: FFmpeg user questions <ffmpeg-user@ffmpeg.org>
Subject: Re: [FFmpeg-user] Leveraging TensorFlow and FFMPEG
You should add " "or ' ' to model path, because there is special character
in model path that breaks command line.
And you should also set input output name:
"dnn_backend=tensorflow:model=$MODEL_PATH:input=input_port_name:output=outpu
t_port_name "
Please pay attention to the data format, ffmpeg-dnn tensorflow backend
passes data in format (RGB, NHWC and range(1.0f ~ 0.0f)) to tensorflow
model.
I am not sure if .h5 model is supported since I only tried .pb model.
Best regards
Wenbin
> Hello All,
>
>
>
> I am attempting to leverage Super-resolution from TensorFlow to
> enhance a video.
>
> I used FFMPEG to split out each frame. 200K of them.
>
> I then installed TensorFlow into my windows 11 pro system and verifies
> the install and path.
>
> All good.
>
>
>
> I downloaded a pre-trained model from Model Zoo. Not sure if I even
> did this right as this is all very new and very complicated. If
> anyone has experience with this stuff and can make a recommendation,
> that would be awesome!!.
>
>
>
> I then wrote a simple script that should enhance the individual frames.
>
>
>
> Script Start:
>
>
>
> echo "Applying super resolution (SRGAN) to each frame..."
>
> $MODEL_PATH =
> "C:\Super_Resolution\weights-srgan\weights\srgan\pre_generator.h5"
>
> $OTHER_OPTIONS = ""
>
> $files = Get-ChildItem *.png
>
> $total_files = $files.Count
>
> $current_file = 0
>
> foreach ($file in $files) {
>
> $current_file++
>
> $progress = [math]::Round(($current_file / $total_files) * 100)
>
> echo "Processing file: $($file.Name) [$current_file/$total_files]
> ($progress%)"
>
> ffmpeg -i "$file" -vf
> dnn_processing=dnn_backend=tensorflow:model=$MODEL_PATH
> $OTHER_OPTIONS
> "enhanced_$($file.Name)"
>
> }
>
>
>
> Script End:
>
>
>
> It is generating this error:
>
>
>
> Input #0, png_pipe, from
> 'F:\Movies\Wholly_Moses\Temp_Work\frame_0001.png':
>
> Duration: N/A, bitrate: N/A
>
> Stream #0:0: Video: png, rgb24(pc, gbr/bt709/bt709), 1920x816 [SAR
> 1:1 DAR 40:17], 25 fps, 25 tbr, 25 tbn
>
> [AVFilterGraph @ 000001c6d46bbc00] No option name near
> 'Super_Resolutionweights-srganweightssrganpre_generator.h5'
>
> [AVFilterGraph @ 000001c6d46bbc00] Error parsing a filter description
> around:
>
> [AVFilterGraph @ 000001c6d46bbc00] Error parsing filterchain
> 'dnn_processing=dnn_backend=tensorflow:model=C:\Super_Resolution\weig
> hts-srg
> an\weights\srgan\pre_generator.h5' around:
>
> [vost#0:0/png @ 000001c6d46bcac0] Error initializing a simple
> filtergraph
>
> Error opening output file enhanced_frame_0001.png.
>
> Error opening output files: Invalid argument
>
>
>
> I have altered the parameters for
> "dnn_processing=dnn_backend=tensorflow" a dozen different ways and it
> still generates the same error.
>
>
>
> Would anyone be able to help me format a correct command syntax?
>
>
>
> Please let me know if there is any other information needed to help
> answer this question.
>
>
>
> Regards,
>
> -Ron
>
> _______________________________________________
> ffmpeg-user mailing list
> ffmpeg-user@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-user
>
> To unsubscribe, visit link above, or email
> ffmpeg-user-requ...@ffmpeg.org with subject "unsubscribe".
_______________________________________________
ffmpeg-user mailing list
ffmpeg-user@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-user
To unsubscribe, visit link above, or email ffmpeg-user-requ...@ffmpeg.org
with subject "unsubscribe".
_______________________________________________
ffmpeg-user mailing list
ffmpeg-user@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-user
To unsubscribe, visit link above, or email
ffmpeg-user-requ...@ffmpeg.org with subject "unsubscribe".
