Timeline for answer to Negative dimension size caused by subtracting 3 from 1 for 'conv2d_2/convolution' by charelf
Current License: CC BY-SA 4.0
Post Revisions
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 15, 2024 at 10:34 | comment | added | charelf | i.postimg.cc/L5qTXTfP/image.jpg | |
| Jan 15, 2024 at 10:30 | comment | added | charelf | Its been a long time since ive done this, so bear with me: If e.g. your base is 4x4, and your convolution layer is 3x3. Try drawing the different positions you can put your conv layer on top of the 4x4 base layer. There are 4 positions if im right. So your output of this operations are a 2x2 new layer. In other words, the convolution operation decreased the size of the base layer. A similar thing happens in my above example. | |
| Jan 14, 2024 at 13:52 | comment | added | BanikPyco | Great intuitive explanation. However, one thing I don't get from your answer is why the reduction from 32x32 to 30x30 after the convolution layer. The division of pixels / pool_size ("window" size) is obvious and can't understand how I did not figured that out myself. The output of convolution layer is: IMG_HEIGHT - kernel_size + 1. | |
| May 19, 2021 at 12:59 | history | edited | charelf | CC BY-SA 4.0 |
added 103 characters in body
|
| Jun 20, 2020 at 9:12 | history | edited | CommunityBot |
Commonmark migration
|
|
| Nov 9, 2019 at 9:08 | history | edited | charelf | CC BY-SA 4.0 |
formatted the text and changed some phrasing and words I did not like
|
| Mar 11, 2018 at 14:40 | review | Late answers | |||
| Mar 11, 2018 at 14:41 | |||||
| Mar 11, 2018 at 14:20 | review | First posts | |||
| Mar 11, 2018 at 14:24 | |||||
| Mar 11, 2018 at 14:18 | history | answered | charelf | CC BY-SA 3.0 |
