Segmenting Satellite Imagery with the Segment Anything Model (SAM)
Вставка
- Опубліковано 19 сер 2024
- Notebook: samgeo.gishub....
leafmap homepage: leafmap.org
geemap homepage: geemap.org
GitHub: github.com/gis...
Twitter: / giswqs
UA-cam channel: / @giswqs
GEE Tutorials Playlist: gishub.org/you...
Leafmap Tutorials Playlist: gishub.org/you...
Geographic Software Design Playlist: gishub.org/gsd
Spatial Data Management Playlist: gishub.org/sdm
Dear Dr. Qiusheng Wu,
I hope this message finds you in good health and high spirits. I am writing to express my gratitude for the insightful and educative content you share on your UA-cam channel. I have found it incredibly helpful in my studies and professional development.
In line with the fascinating work you produce, I'd like to kindly propose a topic for your future content - Pan Sharpening Satellite Images Using Deep Learning. This advanced technique has piqued my interest and I believe an instructional video by an expert such as yourself would be invaluable for me and many other enthusiasts in this field.
While there are resources available on pan sharpening techniques, the focus on deep learning approaches is somewhat limited. Given the transformational role deep learning has been playing in various fields, understanding its application in satellite image enhancement would be greatly beneficial.
I am particularly interested in understanding the best practices, potential challenges, and key points to keep in mind while employing deep learning techniques in the pan sharpening process.
I appreciate your consideration of my request and eagerly look forward to potentially seeing this topic covered on your UA-cam channel. I am certain that a video from you on this subject will contribute significantly to the learning of your viewers, including myself.
Thank you once again for the incredible work you do in educating and inspiring us.
Best regards,
Muito bom! Poderia gravar mais vídeos sobre a temática. Como por exemplo, como adaptar o código para inserir uma imagem de Satélite. tif ao invés de basemap. O plug-in Geometric Attributes no QGIS também trabalha com o SAM, poderia fazer um passo a passo mostrando como escolher os limiares no plug-in, me pareceram confusos. Finalmente, gostaria de parabenizá-lo pelos conteúdos postados e pelo compartilhamento de ideias, ferramentas e conhecimentos.
Amazing, I can't wait for the next model!
This is absolutely brilliant! I was working on field delineation with machine learning and the results from traditional CV algorithms were subpar. I am confident that this will be a great tool. Thank you.
Did it work well ?
@@shidharthrouth it does a very good job if your coordinates if for a field region and you can even get shape files out of it. I said field region because if you don’t have that, it can generalize clustered buildings and the mask will look like a field
can I know which algorithm u have used from ML?
吴老师厉害，感谢老师，只是全英文讲解，对于我来讲有点吃力！
High quality informations ! thanks again for this work !!
Thank you for this great tutorial.
I hope you will continue to work on development and keep us updated with this kind of long videos that explain a lot about (SAM).
I have a question :
What are the possibilities of applying this model to data such as SAR?
Thanks sir. This is great contribution
very great work and nice tutorial👍
Amazing ! Thank you so much :)
A very good job! It's helpful!
感谢你的工作😀
Fascinating ❤️
THANK YOU!
Sir, can you tell me the exact name/source of the downloaded satellite data? Actually, I using it for publication.
谢谢吴老师
This is wonderful! I was wondering if it works to make a footprint out of buildings with this tool.
Has anyone try identifying solar panels? I did it but it didn’t pick it up and only detected a few roofs instead
Thank you laoshi, but can you make a video to show how to convert the results of SAM model from images to excel file?
Sometimes localtileserver conflicts with samgeo and doesn't add the raster properly above the basemap.
Congratulations on your work, it is a great contribution to the gis community. I would like to know if it is possible to download the segment in shp, geojos or some other format to open it in a desktop GIs.
Thank you very much, greetings from Uruguay
Can someone explain, please, in the sense of getting such a segmentation, where we cannot label, but it segments everything in a row? Thanks for tutorial
Hi sir the video regarding SAM is so good and useful for getting the segmentations from the image. I have one query is it possible to extract the features separately like (trees, buildings, land, etc). ThankYou.
Is it possible to get same color for similar objects....for eg one color for all the trees and so on
Thank you very much for sharing this work with us, can you please tel me how did you select the default SAM hyperparamter ? Is it subjective or you did some hyperparamter tunning? if the last one on which data did you tuned it?
I'm asking this question in order to know if this default value are ok to you SamGeo for crops Land boundary detection, or I should try to tweak the hyperparamters. Any recommendation on that would be very appreciated. Thanks again
Great video, I would like to ask something basic. Is it possible to segment aerial images with one class like buildings etc. ? Images are .jpg, I am looking for something exactly like this but I guess there is something that I do not understand.
Yes, it is possible. Stay tuned for the next video
@@giswqsThank you so much, can not wait for it :)
Hi, I read the documnet, there is a example about swimming pool. i also tried setting the mask as trees, and it was successful to mark the area containing trees. So my question is do you set up other labels to mark? like grass, shrubs? Thank you, your videos and documents are really helpful!
Great tutorial sir, Where the output result stored
Under the working directory
Dear Qiusheng, thanks a lot for the tutorial, I have been following your videos on Segmentation using SAM, do you know how we can assign a class label to each segmented object in the image?
Use input promots. samgeo.gishub.org/examples/input_prompts/
Hi professor when I downloaded and import your Jupiter into my vscode ‘ no module import samgeo then I opened conda into my env to install pop up a error ‘ no matching distribution found for samgeo ‘ and ‘could not find a version that satisfies the requirement samgeo’ how can I solve and succeed to install in my own computer thanks
Fantastic tutorial. Is there a way to segment Uint16 images? I have tried the code and it only allows segmentation of Uint8 data type, but that pixel depth is too low for the Sentinel-2 imagery I am using
Hello, I was using the algorithm and I noticed that when I use the polygon to highlight an area of interest, the resulting mask extends to regions that were not selected. Could you please provide any suggestions to fix that? Thank you
Hello Dr.
How can I access the raw data returned from SAM(list of dictionaries that contain information about the generated masks). In the package file in the generate() function I noticed it's initially assigned to self.mask but later self.mask gets overwritten in save_mask() function that is called withing generate(). And in the case that output in not passed when calling generate() and default is None, I keep getting an error.
that's awsome! is it included in the geospatial package?
It is included in the geospatial-ml package. See github.com/opengeos/geospatial-ml
Whoever created this... Who you people come up with this things outa nowhere
新手想問，你能否製作一個本地端的.py，不要用colab方式
不知道可不可以!?
Is possible to add the label of each segmentation?
Does it work to detect crop rows?
Hi professor does this can deal with jp2 in addition to tif thanks
You can read jp2 file as a numpy array using rasterio. The samgeo package accepts numpy array as input.
import rasterio
# Open the JP2 file using rasterio
with rasterio.open('path/to/your/jp2/file.jp2') as dataset:
# Read the image data as a numpy array
jp2_data = dataset.read()
# Convert the numpy array to a standard format
jp2_data = jp2_data.squeeze() # remove the band dimension if it is 1
jp2_data = jp2_data.transpose((1, 2, 0)) # rearrange dimensions from (bands, rows, cols) to (rows, cols, bands)
# Get information about the image size and number of bands
print(f"Image shape: {jp2_data.shape}")
print(f"Number of bands: {jp2_data.shape[2]}")
print(f"Data type: {jp2_data.dtype}")
@@giswqsthis program is automatically segmented? Automatically labeled? Does not need manually labeled? I am curiously really many thanks!
Can we use this model for land cover classes?
Not right now, but potentially in the future
@@giswqs, thanks prof.