I have 2 questions. How are the 1X1 and 3X3 CNN used trained to obtain the weight parameters? Also shouldn't 3X3 with stride 1 change the dimension, though it keeps the number of channels the same the size of the output feature would have changed and reduced by 2
How is this different from U-net? I think they're pretty similar if you think that in the U-net you're going down in the encoder, up in the decoder and sideways with the skip connections. It's like an upside-down U-net
I don't know if I got this wrong but if I take a 1x64x26x26 feature through a convolution that has a K=3 and S=1, I will definitely not end up with a 1x64x26x26, but with a 1x64x24x24. To achieve the desired shape would require a P=1. If I'm not correct, would someone please explain how the dimensions would work in this case?
Though I understand the theory it’s just that I have never implemented/used them myself. I prefer to share those concepts that I have implemented myself and applied on some real world problem. But not saying no :) maybe one day. Thanks for the ask though.
This is quite informative and helpful. Can you please create a video on prediction heads in fpn as in how to assign a predicted bbox to a particular feature map. That would be quite helpful.
Yes, thinking to make some videos about different label assignment techniques. Now about your question - the right terminology or phrasing of your request would be how to assign an anchor box to a particular feature map.
Instead of doing the upsampling via pytorch module and being angry about it, would it be any more useful to train an additional layer to do the upsampling instead? I'm thinking of a layer analogous to the decoder layer in an autoencoder.
No need to be angry at it :) … yes you could do that. As a matter of fact the additional layers after upsampling is to reduce it effects. The cost would be number of parameters. So it is always a trade off.
Keep the pearls of wisdom dropping sir..Privilage to learn from you miles across...
🙏 thanks for the kind words.
very helpful! I really like that you're explaining it with an example with concrete numbers!
🙏
Incredible explanatory skills!
Sir, I have a lot of to say after finding your video on RUclips but just ❤ , respect and thank you. 🙏🙏
🙏
Thank you for sharing your knowledge!
🙏
Excellent tutorial. Thank you very much.
🙏
I am so happy I found this video. Really good content!
🙏
This is excellent! I just love it.
🙏
Thanks for sharing the knowledge
I have 2 questions. How are the 1X1 and 3X3 CNN used trained to obtain the weight parameters? Also shouldn't 3X3 with stride 1 change the dimension, though it keeps the number of channels the same the size of the output feature would have changed and reduced by 2
is useful to add channel and spatial attention in conv layers to improve
amazing explanation Dr.
🙏
How is this different from U-net? I think they're pretty similar if you think that in the U-net you're going down in the encoder, up in the decoder and sideways with the skip connections. It's like an upside-down U-net
I like your videos, which are easy and fun to learn. Thanks a lot!
🙏
If done with UNet, it won't require upsampling as we concatenate the layers right?
I don't know if I got this wrong but if I take a 1x64x26x26 feature through a convolution that has a K=3 and S=1, I will definitely not end up with a 1x64x26x26, but with a 1x64x24x24. To achieve the desired shape would require a P=1.
If I'm not correct, would someone please explain how the dimensions would work in this case?
Could you give a tutorial of diffusing model to your VAE series? Its related and would like to see your explanation!
Though I understand the theory it’s just that I have never implemented/used them myself. I prefer to share those concepts that I have implemented myself and applied on some real world problem.
But not saying no :) maybe one day. Thanks for the ask though.
Do you know how to combine AFPN with the YOLO v8 algorithm? If you know, please tell me. Thanks
Excellent
Thank you... excellent clarity... please try to make a tutorial on anchor free detectors like FCOS..
🙏 yup. First need to implement it :)
Could you give a tutorial on the vision transformer model for object detection?
in some time. have been preoccupied with some stuff but would try my best
what about height and width are odd number (415), sir? In that case, the size after conv and after upsample is miss match. How to fix that, please!
Resize the image to 416 or any other size (e.g. 640) before feeding it to the network.
This is quite informative and helpful. Can you please create a video on prediction heads in fpn as in how to assign a predicted bbox to a particular feature map. That would be quite helpful.
Yes, thinking to make some videos about different label assignment techniques.
Now about your question - the right terminology or phrasing of your request would be how to assign an anchor box to a particular feature map.
Thanks a lot! would be the following videos soon?
🙏 yes.
Instead of doing the upsampling via pytorch module and being angry about it, would it be any more useful to train an additional layer to do the upsampling instead? I'm thinking of a layer analogous to the decoder layer in an autoencoder.
No need to be angry at it :) … yes you could do that. As a matter of fact the additional layers after upsampling is to reduce it effects. The cost would be number of parameters. So it is always a trade off.
@@KapilSachdeva Thank you! informative video btw
🙏
thankyou sir !
🙏
thank you for the content , next video soon?
🙏 … yes. Most likely tomorrow. Thanks for keeping me accountable.
@@KapilSachdeva thank you again for the content, looking forward for more of these videos
Still working on the next video; not yet happy with it hence not published yet.
Hernandez Betty Lewis Kenneth Gonzalez Christopher
Clement Landing
new video when ?
today ... very late sorry :(
Pagac Road
Franco Neck
Can you make the video in Urdu language
There are urdu subtitles and may be that will be of some help!
Thompson Cynthia Martin Frank Brown Jason
Moore Kevin Moore Sharon Lewis Richard
8831 Osvaldo Heights