Timeline for 1x1 Convolution. How does the math work?
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Sep 26, 2018 at 17:59 | vote | accept | Mihkel L. | ||
| Sep 26, 2018 at 17:59 | vote | accept | Mihkel L. | ||
| Sep 26, 2018 at 17:59 | |||||
| Sep 24, 2018 at 17:48 | comment | added | Francesco Pegoraro | Do you still have problems understanding? | |
| Sep 23, 2018 at 22:42 | history | edited | Francesco Pegoraro | CC BY-SA 4.0 |
Add code for example in numpy and magnificent illustrations
|
| Sep 23, 2018 at 22:17 | comment | added | Francesco Pegoraro | I am going to modify my answer with an example in numpy | |
| Sep 23, 2018 at 18:40 | comment | added | Mihkel L. |
best thing would be to show PyTorch code that compiles W x W x K matrix into W x W matrix :)
|
|
| Sep 23, 2018 at 18:30 | comment | added | Mihkel L. | @Francesco Pegoraro Show me some real calculation. I have matrix 28x28x192. When I multiply it with 1x1x192 then i get 28x28x192. No Scalar. Show me how I get scalar please. | |
| Sep 23, 2018 at 14:07 | comment | added | Francesco Pegoraro | You get the shrinking because the result of convolution between overlapping regions between a tensor of depth k and a filter of depth k is a matrix with depth 1! For each region you get a scalar, and all together they form a matrix. So you shrink from k to 1! To obtain a result of depth X, you just stack the result of X filters. | |
| Sep 23, 2018 at 7:46 | comment | added | Mihkel L. |
I'm trying to achive this with matrix multiplication. And there is noway I can get to 28x28xK what's the calculation that shrinks it? If you look at @André answer, then what's g?
|
|
| Sep 22, 2018 at 20:32 | history | edited | Francesco Pegoraro | CC BY-SA 4.0 |
added 1 character in body
|
| Sep 22, 2018 at 15:34 | history | answered | Francesco Pegoraro | CC BY-SA 4.0 |