You need to create an integrated version of that with the image segmentation editor that is already inside the Label Studio, similar to what Labelbox has. Having to add to the backend with additional code that needs to be update or edited all the time creates huge problems in terms of adoption of the platform for all users
Thank you for such a helpful video! Please tell me, if I need to count a lot of small items in one photo, do I need to slice the photo into parts or label all the items I am interested in on the whole photo?
I have used a DAY for this and this is not working. It is simply not ready yet. The problem is that it dosent work when images are on s3. The issues on the github are not being answered. You need to be clear that this ONLY works when having the images locally. That said label studio is cool
Hi, can you please let me know if you found an answer elsewhere ? I will need to export my data in coco format and did mot find if it it possible with this. Thanks
@@corentindoche5645 I havent yet. there were various tools on github that converted brush to polygon but i didnt have much time spare for this as its just a hobby so havent tried anything else yet.
@@rifaiershad At the end I did note use SAM for my project so I cannot give you a 100% sure answer but from what I had read (and understand), it is possible to export it in coco format
This is truly an amazing feature with so much potential, but other than showing off a demo this isn't usable to practically annotate a project. There is no way to actually get usable segmentation masks using any of the export options. Documentation says to export masks via the PNG or Numpy options. I think this is really an issue with Label Studio since this affects exporting masks without Segment Anything, but both of these options do not offer a practical way to annotate data. This is due to the fact there's no way to determine which mask belongs to which image. You could manually go back to label studio and look at the image, find the annotation ID in label studio UI, and rename the mask file accordingly. Again, this works to show somebody in a demo, but would definitely not be practical to label a real project. All of the other export options including YOLO, Coco, and Pascal just export annotations with empty segmentation information.
Man you guys are truly a waste of time. So much talking and still no way to export polygons to make this usable. Instead of making 5000 blog posts try making something useful.
In the video it shown that we can export it in coco format, what's in then ? Why it is not usable ? I am struggling to install the back end at the moment so I can't try on my own, I would be glad if you could help me so I stop wasting my time trying to install it if it is not usable. Thanks
Edit my previous answer: the Export -> COCO in the UI does not export any of the predicted brush annotations that were created with SAM. Therefore, the whole predictions are useless, unless you find a way to convert the brush data to COCO polygons... @@corentindoche5645
You need to create an integrated version of that with the image segmentation editor that is already inside the Label Studio, similar to what Labelbox has. Having to add to the backend with additional code that needs to be update or edited all the time creates huge problems in terms of adoption of the platform for all users
i wish man💯 i spent 5hrs setting it up now the model wouldn't work....am just going back to Roboflow
Thank you for such a helpful video!
Please tell me, if I need to count a lot of small items in one photo, do I need to slice the photo into parts or label all the items I am interested in on the whole photo?
Great, thanks a lot
Hi, i am having some issue to build the docker following your instructions in the video and tutorial...is it possible to have any help?
YOU KNOW.... YOU KNOW
.. YOU KNOW...
label studio save raster mask or a closed polygon?
Are there any opportunities for data annotator ?
I'm looking for full time job ( Remote ) or freelance
This tool would be successful with polygons to fix IA errors
I MEAAAN UUUUMMMMMMMMMMM 💅💅💅
I have used a DAY for this and this is not working. It is simply not ready yet. The problem is that it dosent work when images are on s3. The issues on the github are not being answered. You need to be clear that this ONLY works when having the images locally.
That said label studio is cool
this guy said you know at least 300 times
uhhhhh uhhhhh uhhhh XD. not to be mean, but this was a good video and he is good, but it became annoying
create a model to go through the video audio and cut the you know and uhh parts
any way for this to work with polygons instead of brush so its suitable for coco?
Hi, can you please let me know if you found an answer elsewhere ? I will need to export my data in coco format and did mot find if it it possible with this.
Thanks
@@corentindoche5645 I havent yet. there were various tools on github that converted brush to polygon but i didnt have much time spare for this as its just a hobby so havent tried anything else yet.
@@corentindoche5645 any update on this?
@@rifaiershad At the end I did note use SAM for my project so I cannot give you a 100% sure answer but from what I had read (and understand), it is possible to export it in coco format
@@corentindoche5645 references on exporting it to coco? thanks btw
This is truly an amazing feature with so much potential, but other than showing off a demo this isn't usable to practically annotate a project.
There is no way to actually get usable segmentation masks using any of the export options. Documentation says to export masks via the PNG or Numpy options. I think this is really an issue with Label Studio since this affects exporting masks without Segment Anything, but both of these options do not offer a practical way to annotate data. This is due to the fact there's no way to determine which mask belongs to which image. You could manually go back to label studio and look at the image, find the annotation ID in label studio UI, and rename the mask file accordingly. Again, this works to show somebody in a demo, but would definitely not be practical to label a real project.
All of the other export options including YOLO, Coco, and Pascal just export annotations with empty segmentation information.
i am a data labeler from Ghana, looking for a remote job opportunity, help me
Man you guys are truly a waste of time. So much talking and still no way to export polygons to make this usable. Instead of making 5000 blog posts try making something useful.
In the video it shown that we can export it in coco format, what's in then ? Why it is not usable ? I am struggling to install the back end at the moment so I can't try on my own, I would be glad if you could help me so I stop wasting my time trying to install it if it is not usable. Thanks
Edit my previous answer: the Export -> COCO in the UI does not export any of the predicted brush annotations that were created with SAM. Therefore, the whole predictions are useless, unless you find a way to convert the brush data to COCO polygons... @@corentindoche5645
@@corentindoche5645 Did u installed it, bro?
@@corentindoche5645same with me. I could not build sam image in docker