Meta New Segmentation Model
ai.facebook.com
Meta New Segmentation Model
1–10 of 14 posts
Re: Meta New Segmentation Model
#2Here it is: https://github.com/facebookresearch/segment-anything#model-c...
Licensed under Apache 2.0 license, for free.
Go for it.
Re: Meta New Segmentation Model
#3Re: Meta New Segmentation Model
#4Its funny to me that they are able to produce cool R&D stuff like this, but they chose to go all in on the metaverse.
Re: Meta New Segmentation Model
#5very impressive work by Meta AI
Re: Meta New Segmentation Model
#6Its funny to me that they are able to produce cool R&D stuff like this, but they chose to go all in on the metaverse.
Re: Meta New Segmentation Model
#7Initial impression: it seems to be pretty good at identifying objects but when you look at the cutouts without the glowing outline you'll quickly realize the edge is rather rough and thinner parts sometimes get omitted completely. This might be though because those borders they draw are too thick and omitted from the resulting cutout.
Example: https://i.imgur.com/S57c5Cj.png
In the kids drawing both extracted person had no arms.
Re: Meta New Segmentation Model
#8Its funny to me that they are able to produce cool R&D stuff like this, but they chose to go all in on the metaverse.
The metaverse application of image segmentation was shown in the GIF in the article.
There would be no need for image segmentation in the metaverse, a completely generated construct. The information gained from image segmentation could be derived much more easily and embedded directly in the objects you're interacting with.
Re: Meta New Segmentation Model
#9Segment Anything Model (SAM) can "cut out" any object in an image - https://news.ycombinator.com/item?id=35455566 - April 2023 (33 comments)
Re: Meta New Segmentation Model
#10Earlier quoted context omitted.
The metaverse application of image segmentation was shown in the GIF in the article.
Are you confusing the metaverse with AR/VR in general? There would be no need for image segmentation in the metaverse, a completely generated construct. The information gained from image segmentation could be derived much more easily and embedded directly in the objects you're interacting with.