facebookresearch
Not grabbing entire drawing...
Hi,
so I starting from scratch here... on a clean new Mac and have insalled Anaconda and Docker - there goes 8GB of space :-) and following your excellent instructions have got things running. I have used python before but I'm only a hobby level coder.
I want to use my daughters drawings - the first one worked fabulously. The second however gives many of these errors as it runs.
point [[0.44142616 0.13921902]] not inside or on edge of any triangle in mesh. Skipping it
and then the result gif is like this...
... or when I upped the gutter on the image, making the canvas bigger...
The generated mask image is way off...
I tried using annotations_to_animation.py but the image was already clipped at the edges.
Do you have any tips for stylistic changes to the image which would make the image detection more reliable ( I tried adding a black stroke to the image and got this)... but it still clips...
Are there optimum dimensions or resolution? I tried ensuring alpha transparency in the png...
Aha... I think I've got it... I made the picture much bigger ... 3000+ pixel with the image (small) in the centre.
THEN I got this...
...which is much better but it's still clipping slightly... and when I try and fix_annotations I see this...
Our plan is to generate sprite sheets (or folders of numbered images) so she can make a game with them, so I assume I can modify the motion params (somewhere) for "walking", for example, so that the character doesn't move.
Lastly, that slight white border might get a bit annoying (given the image itself had no background)... looks at bit like characters from Paddington Bear when I was kid https://youtu.be/mClA14WQFu8?t=55 ) but we will be shrinking the generated image down a lot so it may get lost in the shrinkage...
Thanks!
Hello,
If the tools for automatically creating the texture.png, mask.png, and joints aren't working, you can also create them by hand. texture.png should be a cropped RGB image. mask.png should be a grayscale image with the same dimensions as texture.png. And you can reuse an existing char_config.yaml from your previous attempts. Run it through the annotation fixer to make sure it looks good.
OK, so wow! That has made a huge difference...
Can I kind of, by way of a workflow, replace the texture.png and mask.png with any images (same size) and then run fix_annotations.py to tweak the skelton, then image_to_animation.py? to get a gif?
Continuing, or given that ideally I want to generate idle, walk, dance, dab animations for any given character image, will it be better to duplicate and adapt image_to_animation.py (kind of as shown in the documentation above)?
This is great fun!
Glad you're enjoying the tool :)
Yes, you can replace texture.png and mask.png with whatever you want and run fix_annotations.py to tweak the skeleton. The files that result from that should be good to go, and you can create an animation from them.