I am following the FCN for semantic segmentation (describe in
https://devblogs.nvidia.com/parallelforall/image-segmentation-using-digits-5/).
For fcn_alexnet, the crop offset size is 18 by calculated backward from last layer.
But I don't understand how I do get the offset value, 5 and 9 in score_pool4c and score_pool3c layers respectively as below
It would be very thankful if somebody help me to how to calculate the below offset number.
Thank you
-Joshua
...
layer { |
| name: "score_pool4c" |
| type: "Crop" |
| bottom: "score_pool4" |
| bottom: "upscore2" |
| top: "score_pool4c" |
| crop_param { |
| axis: 2 |
| offset: 5 |
| } |
| }
...
| layer { | | name: "score_pool3c" | | type: "Crop" | | bottom: "score_pool3" | | bottom: "upscore_pool4" | | top: "score_pool3c" | | crop_param { | | axis: 2 | | offset: 9 | | } | | } |
|