Another great presentation. I find it fascinating that reformulating the output from a binary mask to flows results in such performance. Do you think this yields any insight on how (or what) the U-net learns?
By representing the output of the U-net as flows, Cellpose is able to capture more information about the spatial relationships between cells. This information is then used to improve the accuracy of the segmentation algorithm. The fact that reformulating the output of the U-net as flows results in such performance suggests that the U-net is learning to represent the spatial relationships between cells in a way that is more than just a binary mask. This is an important insight, as it suggests that the U-net is learning more than just the boundaries of cells.
Hi Sreeni, great great video thanks a lot, I'm going to use this model to my pap smear dataset, right now with stardist I have decent results but not great in segmenting multiclass, do you recomend another model?, so far I have good results in the Apeer Zeiss but unfortunately the model is closed and I cannot see the strecture used.
Hi, can you post some example images? There are several options you can try - SplineDist, DCSA-Net, lacss, torch-em, MaskRCNN, detectron2 and many more.
@@kulkarniajinkya Hi, sorry TH-cam delete my comment every time I put a link to my database, you can look in google: CRIC database, it´s a classification database of Cervix pap smear, I already labeled the nuclei in segmentation.
Hi sir, I am a high school student currently doing research using this same dataset and I am having issues with the handcrafted feature extraction method, in terms of segmenting all the images in the dataset I would like to ask for help
I have been experimenting with Detectron2, amazing results (see my LinkedIn posts). I recommend you to try it out. I will try to record a tutorial in the next couple of weeks.
Hi Sreeni. I have advanced a lot in my research thanks to your insightful videos. I keep having problems with segmentation. Algorithms like cellpose or others you have explored in the channel work very well for circular or oval objects, or for those that border each other. I study glial cells in the brain, objects that are highly irregular, branching, and strongly overlapping with each other. It would be great if you could explore some object segmentation and identification algorithms for those cases. Greetings from Canada
Glial cells (e.g., Astrocytes) are challenging to segment as individual objects. I've read a couple of papers on the topic but haven't found a good approach. The best approach that kind of works is to performing semantic segmentation (e.g., using U-net) followed by watershed. You may find traditional techniques to work better, such as Random Walker segmentation. Here is an interesting article on this topic: www.ncbi.nlm.nih.gov/pmc/articles/PMC9396514/
Hello, did you mention changing the Python version in Google Collab in your training? Thank you if there is a way to do this For example, changing the current version 3.10.6 to 3.6
To change python version in colab: First check the current version: !python --version Install the desired version: !apt-get install python3.6 Create symbolic link to the new version: !ln -sf /usr/bin/python3.6 /usr/bin/python3 Now check the version agaon to confirm that the new version is being used: !python --version
Hello 👋! Thank you for video 😊, all your videos are very helpful❤. Could you help me, I need to choose best PC for machine learning especially for segmentation task
Hi Sreeni! Thanks fo the video! I am currently a high school student doing research on cervical cancer and I am evaluating the performance of cervical cancer classification systems specifically using handcrafted features vs CNN features. For the handcrafted features, I am having issues with my code and would like to ask whether you could be of help in that. I can share my linkedln/email if needed. I would really appreciate it thank you!
![[UNCUT] "บอสณวัฒน์" แฉ! แหล่งซุกเงินแก๊งบอส ขยันผิดที่ไม่กี่ปีก็เข้าคุก l คนดังนั่งเคลียร์ l17ต.ค.67](http://i.ytimg.com/vi/BL56qIfXYZM/mqdefault.jpg)
Thank You Alot. You are the best teacher.
Another great presentation. I find it fascinating that reformulating the output from a binary mask to flows results in such performance. Do you think this yields any insight on how (or what) the U-net learns?
By representing the output of the U-net as flows, Cellpose is able to capture more information about the spatial relationships between cells. This information is then used to improve the accuracy of the segmentation algorithm.
The fact that reformulating the output of the U-net as flows results in such performance suggests that the U-net is learning to represent the spatial relationships between cells in a way that is more than just a binary mask. This is an important insight, as it suggests that the U-net is learning more than just the boundaries of cells.
Hi Sreeni, great great video thanks a lot, I'm going to use this model to my pap smear dataset, right now with stardist I have decent results but not great in segmenting multiclass, do you recomend another model?, so far I have good results in the Apeer Zeiss but unfortunately the model is closed and I cannot see the strecture used.
Hi, can you post some example images? There are several options you can try - SplineDist, DCSA-Net, lacss, torch-em, MaskRCNN, detectron2 and many more.
@@kulkarniajinkya Hi, sorry TH-cam delete my comment every time I put a link to my database, you can look in google: CRIC database, it´s a classification database of Cervix pap smear, I already labeled the nuclei in segmentation.
Hi sir, I am a high school student currently doing research using this same dataset and I am having issues with the handcrafted feature extraction method, in terms of segmenting all the images in the dataset I would like to ask for help
I have been experimenting with Detectron2, amazing results (see my LinkedIn posts). I recommend you to try it out. I will try to record a tutorial in the next couple of weeks.
@@DigitalSreeni Thanks for the answer, I'll check it and am anxious to see your video
Hi Sreeni. I have advanced a lot in my research thanks to your insightful videos. I keep having problems with segmentation. Algorithms like cellpose or others you have explored in the channel work very well for circular or oval objects, or for those that border each other. I study glial cells in the brain, objects that are highly irregular, branching, and strongly overlapping with each other. It would be great if you could explore some object segmentation and identification algorithms for those cases. Greetings from Canada
Glial cells (e.g., Astrocytes) are challenging to segment as individual objects. I've read a couple of papers on the topic but haven't found a good approach. The best approach that kind of works is to performing semantic segmentation (e.g., using U-net) followed by watershed. You may find traditional techniques to work better, such as Random Walker segmentation.
Here is an interesting article on this topic: www.ncbi.nlm.nih.gov/pmc/articles/PMC9396514/
@@DigitalSreeni I appreciate the hint Sreeni. Thanks
@@DigitalSreeni do you think Omnipose might help in such a situation? In any case, it would be helpful if you could make a video on Omnipose.
Hello, did you mention changing the Python version in Google Collab in your training? Thank you if there is a way to do this
For example, changing the current version 3.10.6 to 3.6
To change python version in colab:
First check the current version: !python --version
Install the desired version: !apt-get install python3.6
Create symbolic link to the new version: !ln -sf /usr/bin/python3.6 /usr/bin/python3
Now check the version agaon to confirm that the new version is being used: !python --version
Thank You 😍
Hello 👋! Thank you for video 😊, all your videos are very helpful❤. Could you help me, I need to choose best PC for machine learning especially for segmentation task
Hi Sreeni! Thanks fo the video! I am currently a high school student doing research on cervical cancer and I am evaluating the performance of cervical cancer classification systems specifically using handcrafted features vs CNN features. For the handcrafted features, I am having issues with my code and would like to ask whether you could be of help in that. I can share my linkedln/email if needed. I would really appreciate it thank you!
Hello sir please I need the code for unmasking face by gan network
is cellpose a CNN?
Yes. It uses a U-Net architecture.