[FFmpeg-devel] [PATCH][GSoC] srcnn - an image super resolution filter using CNN

[Mina]

Hi,
   This patch is introduced as a qualification task required by Super 
Resolution project for GSoC. It passes patchcheck and make fate and 
doesn't introduce new warnings. It's implemented by the help of the 
mentor: Pedro Arthur. It's been tested over 7 images of which 6 got 
expected results while 1 didn't.
Used command for testing:
ffmpeg -i input_image -vf "scale=2*iw:2*ih,format=yuv420p,srcnn" 
output_image.

Attached is the patch and here is a link <https://imgur.com/a/pTZTm> for 
3 images with their results; one of which is the one with incorrect result.
Please check(the problem maybe in the way I am copying the image) and 
share some feedback or guidance about what the problem may be and what 
should I do next regarding the GSoC proposal as I haven't submitted one 
yet. Is there any template/requirements for proposal? Also note that the 
algorithm runs convolution without multi-threading on cpu as simple 
while loops so it's slow for large images. Performance be improved later 
on.

Best regards,
Mina Sami.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: srcnn.patch
Type: text/x-patch
Size: 91446 bytes
Desc: not available
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20180322/d0638257/attachment.bin>