Why does the Adaptive Concatenation Module choose 0.5 as the threshold for sigmoid(weight)?

JiayuanWang-JW / YOLOv8-multi-task

GNU Affero General Public License v3.0

235 stars 39 forks source link

Why does the Adaptive Concatenation Module choose 0.5 as the threshold for sigmoid(weight)? #37

Closed TonyMacedonia closed 5 months ago

TonyMacedonia commented 7 months ago

Snipaste_2024-02-26_10-37-19

JiayuanWang-JW commented 5 months ago

As consider this is a threshold. So you can set whatever you want. The learnable parameter is weight instead of this threshold. Maybe after tunning this threshold, the performance may be better than I report. Due to the resource limitation. I didn't based on my work fine-tune the parameters.