wow, it's been a long time, thanks youtube algorithm. I finally saw you again, follow you since Mask-RCNN, but no notifications for your new video. And thanks for sharing this. I have tested your code on any custom image, and SAM is amazing! Love what you doing, keep sharing your knowledge Aarohi! you are the master of applying those new techs to real life.
Hello Aarohi, Great content. Thank you so much! Can you also prepare a video on fine tuning SAM model on custom dataset and how to use it for live inference?
Thanks for your video! Hi, Aaroh. The appearance of SAM makes me feel the cv area has been explored completely. I used to choose cv as my research direction of master degree. But now it shows I should change my choose. What direction of AI area should I learn to get a good job after graduation? Could I get ur recommendations please? Thanks a lot!
While it's true that the computer vision field has seen significant advancements in recent years, there are still many open research problems and opportunities for innovation in this area. Additionally, computer vision is just one of many subfields within artificial intelligence, and there are many other areas that are also experiencing rapid growth and development. Some of these areas include: Natural Language Processing (NLP), Reinforcement Learning (RL), Generative Adversarial Networks (GANs), Autonomous Systems, Healthcare AI. These are just a few examples of the many areas within AI that are currently experiencing rapid growth and development. Ultimately, the choice of research direction should be based on your interests, skills, and career goals.
Hi Aarohi, thanks for sharing these videos. I want to ask you something, is there any microcontroller or device to build a drone that houses AI, then it decides by itself based on a trained model? I found that Raspberry PI Zero 2W is a great candidate, small but no powerful. Now, I found NVidia Jetson Nano, great for AI apps but too big for a drone. Do you happen to know or guide me on this project? Thanks again
I am still exploring the algorithm. Can't answer on this right now. But, I mean this model can segment anything as per Meta. So you can try it on the images which have objects of your custom requirement. If it is working fine on the images of your choice then you can simply import this model.
Yes you are right. If you will check there dataset page, they have mentioned that "NOTE: There are no class labels for the images or mask annotations." We need to figure out the solution for this. I will try and will share if I were able to do it.
I am not certain about the answer to your question as the model is newly developed and I haven't explored it properly yet. But I think in next few days, I can answer that 🙂
Graet video !! Is it possible to automatically segment polygonal or rectangle objects? For example: I have a painting photo which has been taken unprofessionally using a phone. If possible, please make tutorials on how to do so them in python as well as sharing requirements file. It would be great. Thank you in advance!!!
Thanks for this video Aarohi. can you also make a video on the parameters used while doing inference with the pre trained weights. what will be the effect of each parameter. For example if I want to reduce the sensitivity of detected masks then how to do it. For example segment anything detects different mask for the leg, foot, fingers but I need a single mask for all 3 then how to tune it. If I give it a bottle image then it segments the cap and bottle part separately but for now I need single mask.
Meta SAM for design seems very weak, the background removal is very clunky and pixelated, is not smooth at all. Waste of time for design, I'm sure it will be great for other AI uses.
@@NikaTadic Yes, is possible to remove background, but the final result is not good for design purposes as the cut out is not smooth at all. SAM still seems good for AI Vision related things.
wow, it's been a long time, thanks youtube algorithm. I finally saw you again, follow you since Mask-RCNN, but no notifications for your new video. And thanks for sharing this. I have tested your code on any custom image, and SAM is amazing! Love what you doing, keep sharing your knowledge Aarohi! you are the master of applying those new techs to real life.
Glad to hear from you and thanks for your kind words. I have posted videos on new topics. You can check them as well🙂
Great video! A great starting point!!
Thanks so much!
Great explanation and detail. thank you
You are welcome!
Awesome , Thankyou.
You are welcome!
How to generate mask using Text prompts?
how can i take out of data (frame after segment, and location mask to put it in to txt file for trainning yolo?
Hello Aarohi, Great content. Thank you so much!
Can you also prepare a video on fine tuning SAM model on custom dataset and how to use it for live inference?
Nice work, keep it up
Thanks, will do!
Thanks for your video!
Hi, Aaroh. The appearance of SAM makes me feel the cv area has been explored completely. I used to choose cv as my research direction of master degree. But now it shows I should change my choose. What direction of AI area should I learn to get a good job after graduation? Could I get ur recommendations please? Thanks a lot!
While it's true that the computer vision field has seen significant advancements in recent years, there are still many open research problems and opportunities for innovation in this area. Additionally, computer vision is just one of many subfields within artificial intelligence, and there are many other areas that are also experiencing rapid growth and development. Some of these areas include: Natural Language Processing (NLP), Reinforcement Learning (RL), Generative Adversarial Networks (GANs), Autonomous Systems, Healthcare AI.
These are just a few examples of the many areas within AI that are currently experiencing rapid growth and development. Ultimately, the choice of research direction should be based on your interests, skills, and career goals.
Hi Aarohi, thanks for sharing these videos. I want to ask you something, is there any microcontroller or device to build a drone that houses AI, then it decides by itself based on a trained model? I found that Raspberry PI Zero 2W is a great candidate, small but no powerful. Now, I found NVidia Jetson Nano, great for AI apps but too big for a drone. Do you happen to know or guide me on this project? Thanks again
Not that I know of
Instead of using the marker point, how about we use bounding box for the object we want to segment.
That way it wouod be more accurate
Yes, You are correct. Here is the video putting masks using bounding boxes ua-cam.com/video/XB9zg99x2jE/v-deo.html
Hello Aarohi, wondering if I can detect two objects and find their intersection in an image ?
Yes, You can detect 2 objects and do various different operations on it. To suggest you something, I need to know the entire problem statement.
Tried the code today, it is not putting star on truck, can you help ?
I have to check the code to help you with that. YOu can email me at aarohisingla1987@gmail.com
thanks, will do@@CodeWithAarohi
Just now forwarded code. You can see it plot images, but plotting star, is not working. My guess is CUDA is not installed. @@CodeWithAarohi
What is difference with other segment models?
Hi Arohi, is it possible to use it for custom object detection in own dataset?? I mean custom dataset
I am still exploring the algorithm. Can't answer on this right now. But, I mean this model can segment anything as per Meta. So you can try it on the images which have objects of your custom requirement. If it is working fine on the images of your choice then you can simply import this model.
Nice informetion video but can you make more video on this topics as soon as posibal
Yes, sure
Hi Mam. Can we use this model for plant disease segmentation?
Can we retrain this model on the data that we are interested in?
i've look a lot to find a way label the masks , the only way could we get benefit of this SAM ,
ANY ONE KNOW HOW TO LABEL SEGMENTED ELEMENTS ?
Yes you are right. If you will check there dataset page, they have mentioned that
"NOTE: There are no class labels for the images or mask annotations." We need to figure out the solution for this. I will try and will share if I were able to do it.
how this perform in real life application in fps, for example segmenting road
I am not certain about the answer to your question as the model is newly developed and I haven't explored it properly yet. But I think in next few days, I can answer that 🙂
thank you for this wonderful video could please make a video on resnet inception v2
Noted
code pls😢
The official repo of SAM have the code which I used in this video.
Graet video !!
Is it possible to automatically segment polygonal or rectangle objects? For example: I have a painting photo which has been taken unprofessionally using a phone. If possible, please make tutorials on how to do so them in python as well as sharing requirements file. It would be great.
Thank you in advance!!!
I will try to do that in future.
Hi Aarohi , Can we get tutorials on NLP like bert, Transformers...
I will try to make tutorials on requested topics
@@CodeWithAarohi hi Arohi
can you make it full informetion or tutorials video of that topics
Thank you madam
You’re welcome 😊
Thanks for this video Aarohi. can you also make a video on the parameters used while doing inference with the pre trained weights. what will be the effect of each parameter. For example if I want to reduce the sensitivity of detected masks then how to do it. For example segment anything detects different mask for the leg, foot, fingers but I need a single mask for all 3 then how to tune it. If I give it a bottle image then it segments the cap and bottle part separately but for now I need single mask.
I will try
@@CodeWithAarohi Thanks Aarohi. Did you plan anything to make this one ?
@@umairsabir6686 please check this video: ua-cam.com/video/XB9zg99x2jE/v-deo.html
Meta SAM for design seems very weak, the background removal is very clunky and pixelated, is not smooth at all. Waste of time for design, I'm sure it will be great for other AI uses.
Were you able to remove background? I am doing segmenting everything, however I need to segment all objects in the image but disregard the background
@@NikaTadic Yes, is possible to remove background, but the final result is not good for design purposes as the cut out is not smooth at all. SAM still seems good for AI Vision related things.
@@jorgetoro6844 I don’t need it for design purposes so it might be helpful for me. Do you mind to share how did you remove it? Thank you!