307 - Segment your images in python without training using Segment Anything Model (SAM)

CUDA works only on windows and linux apparently, is there a workaround to make it work for Mac, please?

Hello, Thanks for the great video. But I have a question, Can I export the masked output as shp or tiff format?

I am getting CUDA FALSE, sorry but simple question, how to install it ?

Nice one thanks

Thank you for the informative video tutorial! I'm currently working on agriculture land cover monitoring, and I'm wondering if this model can be used for that purpose. Can you confirm if it's suitable for agriculture land cover monitoring?

A video for fine tuning Segement Anything Model would be great

I use a MAC -> Anaconda -> Jupyter notebook. it says CUDA is available: False. 
How can i fix this?.

Can we used this one on segmentation of geotiff file format data?

Hy, sir I am from Geoinformatics background, and I want to get started with Py to process image processing on remote sensed data, where should I start what Gui would you recommend for Py Spyder or vs code is fine i want to visualize the data too while working on it.

I would love to know how they do the hover and onclick segment / mask

Sir, Getting error with sam chcekpoint please suggest how it can be resolved

Hello, what kind of GPU are you using? I want to know if it's possible to do this in real time at at least 1Hz for video.

Hey there ppl! . Is it possible to test the model in CPU? How?.
I got an: CUDA is available: False warning trying to do that.Thx!

your contributions are amazing. wow. thank you so much

Thank you for sharing! By the way, I like your Spyder IDE.

please more videos about this model 🙏

Great tutorial! Thanks for sharing. I would like to create a new cut out image based on the segmentation made by the model. The web demo of SAM already does this and I would like to do the same but using python. Any ideas in how should I attack this problem? Thanks in advance!😄

thanks a lot, it works great for my microscopic data, just very slow segmentation for live monitoring

I don't have a GPU but I have a colab pro plus subscription, can I run this on Colab, can somebody tell me how I can do that please?

Thank u for this excellent video sir. Currently this is doing instance segmentation. Is it possible to alter this code to perform semantic segmentation ?

Hi, could you make a tutorial on how to connect SAM outputs with convolutional networks to determine what object it is or extract information? Regards.
Great video!

Finally omg 😃 Sir please use pytorch

How to save the mask as vector or the masked image ?

promosm 😂

Does this make the other models we discussed on this channel, like UNet, ResNet etc., obsolete? Is it still worth it to learn about them?

Hi Sreeni. Thanks for this lively tutorial of SAM examples. May I know the GPU memory you have? I have run the house.jpg (which is 200KB) and neurons.jpg (which is examples on my CPU (with 16GB RAM) and they worked fine. But when I tried to work with an image that is 25MB size, my system reported as OOM. So, I was wondering if we need a high memory GPU or high RAM CPU to run this for even simple images?

Can you please show how to segment an object using the text prompt

How can ı labelimg data using sam's mask output? Is this possible?

how to measure the area of segmented image??

Wow! I watched your video again and ran through the example you provided. SAM is a great annotation tool and a real labor saver. Fabulous presentation Sreeni.

can you do a video using the text prompt with the code?

How do you apply labels to them? Making a mask is one thing, but how do you label classes?

I'm trying to create tumor segmentator with SAM. When i give mri images to it, it is segments anything in the images absolutely magnificently, but the problem is that it segments all the things and it does not gives any meaningful label. I can't distinguish the tumor masks among all the maskes. I don't know how to do it. I'm open to any suggestion. I apretiace.

how to controle segmented classes like only houses or road and houses only

@DigitalSreeni an example for segmentation with training plz ?:D

To extract labels I did this function maybe can help:

def create_labeled_mask(anns, height, width, max_num_classess):
if len(anns) == 0:
return

sorted_anns = sorted(anns, key=(lambda x: x['area']), reverse=True)
sorted_anns = sorted_anns[:max_num_classess]

label = 1
labeled_mask = np.zeros((height, width))

for ann in sorted_anns:
m = ann['segmentation']
labeled_mask[m > 0] = label
label += 1
labeled_mask = labeled_mask.astype(np.uint8)
return labeled_mask

How does it work on images without dye staining?

Thank you.
Other than the fact that it is open source, what advantage does this toolkit offer over Image-Pro and it’s programming environment for microscopy?

SWEEEEET!!! Thank you for this!

Can you do binary segmentation using SAM?

What a great and timely video. Thank you! I wasn't aware of SAM prior to seeing your presentation.

can we use tiff images?

Thanks!

How to get the labels of the segmentation?

Heyy, thanks Sreeni for your awesome content as always! BTW is there any way I can access a specific portion of mask from the segmentation mask dictionary?? Let's say if I just want to access the mask value where the person is in the image? This could also differ from image to image so I don't know if it works

ViT stands for Vision Transformers. I used UNETR and SwinUNTER (vision transformers) and both of them had excellent performance specifically if your dataset size is very small.

Thanks

I would imagine researchers will want to use SAM to pre-process images for input into other models. But, it seems that SAM takes 20-30 seconds to segment images which is not feasible for thousands of training images. What are your thoughts on how SAM can be used in cases like these? Do you think there will be a way to speed up the segmentation per image?

Thank you for the video

Excellent one superb