ethanhe42 / softer-NMS

Bounding Box Regression with Uncertainty for Accurate Object Detection (CVPR'19)

Home Page:https://github.com/ethanhe42/KL-Loss

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

我在fine turn我之前训练的模型的时候,为什么bbox_pred_std_abs_logw_loss为负值?

gittigxuy opened this issue · comments

commented

我已经将三个都设置为true了

log的原因,这一项可以是负值

但是负值训练绝对loss数值越来越大,从"bbox_pred_std_abs_logw_loss": -0.083365。到。"bbox_pred_std_abs_logw_loss": -1.142597, 而且最后的mAP值降低了,前9万次只将XYXY项设为True时mAP为0.75。然后把所有增加项都设为True后又接着训练到18万次,mAP值却只有0.74???

this repo is deprecated. Please use our new repo which also improved performance: https://github.com/yihui-he/KL-Loss