search

thtrieu / darkflow

Translate darknet to tensorflow. Load trained weights, retrain/fine-tune using tensorflow, export constant graph def to mobile devices
GNU General Public License v3.0
6.13k stars 2.08k forks source link

why loss almost unchanged, but mAP is getting better #1065

Open Once88 opened 5 years ago

Once88 commented 5 years ago

When I train a model, the loss converges quickly after a few steps and almost does not changes later, but the accuracy (mAP) still getting better and better. Can someone help explain this? Thanks a lot.

monoloxo commented 4 years ago

maybe different weights can get almost same loss when it is small,but different weights can get different map