Greg, Its great to be here and to have found your channel. I am trying to do something similar but instead attempt a reconstruction and not classification. As usual I intend to have ResNet architecture consisting of a repeated residual block that consists of 2 convolutional layers and an additive residual connection. The residual block is then repeated for a set amount, e.g. say 10. However, additionally, I intend to treat each residual block as an iteration that outputs one updated reconstruction, then I intend to compute the gradient of the data fit before each residual block of my ResNet. By concatenating current reconstruction and gradient I intend to use this as an input to each subsequent residual block. I can compute the gradient using tape. Thats ok. How do I concatenate the gradient with output from previous residual block and how does that change your code in the notebook? Appreciate your help with learning about ResNets in TF. Thank you!
thanks for the amazing explanation! is it possible to change the skip connections after the training is done of the model? how can one do that with this code what would we call?
Take my courses at mlnow.ai/!
thanks you for this wonderful video ,could please make video on resnet inception v2
Greg, Its great to be here and to have found your channel. I am trying to do something similar but instead attempt a reconstruction and not classification. As usual I intend to have ResNet architecture consisting of a repeated residual block that consists of 2 convolutional layers and an additive residual connection. The residual block is then repeated for a set amount, e.g. say 10. However, additionally, I intend to treat each residual block as an iteration that outputs one updated reconstruction, then I intend to compute the gradient of the data fit before each residual block of my ResNet. By concatenating current reconstruction and gradient I intend to use this as an input to each subsequent residual block. I can compute the gradient using tape. Thats ok. How do I concatenate the gradient with output from previous residual block and how does that change your code in the notebook? Appreciate your help with learning about ResNets in TF. Thank you!
Awesome explanation dude!
Thank you!
thanks for the amazing explanation! is it possible to change the skip connections after the training is done of the model? how can one do that with this code what would we call?
I think it’s important not to say a stride of 1 doesn’t lose information. Convolutions do lose information; dot products are dimensional reducers
Thank you for this!
You're very welcome 😁
Very goood!
thanks brother
Np brother