https://colourise.sg/ pretty cool site setup by GovTech to colourise b&w photos. Just tested some of mine. and yes, once the colourisation is done, you can download it. Sadly, the code is not on gitlab/hub though.
-
-
Replying to @harishpillay
Very cool, I ran quite a few of my old family pics through there and some are very well done. The blog post points to DeOldify as the starting point, not sure from their blog how much they changed the code vs used different training material. https://github.com/jantic/DeOldify
1 reply 0 retweets 0 likes -
Replying to @robbiehonerkamp @harishpillay
We used
@citnaj’s architecture and trained it on Singaporean historical images and Open Images. And so we plan to release our model some time in March, in addition to a CoLab notebook that you can use to colourise your own photos on Google Drive.@GovTechSG1 reply 2 retweets 4 likes -
Replying to @colourisesg @robbiehonerkamp and
Great stuff- what an honor! I am curious- how much of a difference did Open Images make? I've been considering it but it's huuuuuuuge.
2 replies 0 retweets 1 like -
Replying to @citnaj @robbiehonerkamp and
Thanks! We were really inspired by your project. We wanted to use body part-related images from Open Images to improve limb/body colourisation - we used https://github.com/EscVM/OIDv4_ToolKit … to pick out these images by category and removed duplicates across multiple categories.
1 reply 1 retweet 3 likes -
Replying to @colourisesg @citnaj and
We ended up with ~100K body part related images. And while the training didn't really improve colourisation of all limbs, it did help with colourising specific features like the lips/mouth and feet.
1 reply 1 retweet 2 likes
Very interesting to hear. People are the most important part of images to get right I think. Thanks for sharing this!
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.