zhangzhengde0225 / CDNet

The tutorials, datasets and source codes of the crosswalk detection (zebra crossing detection) network, which is robust in real scenes and real-time in Jetson nano. cross. detect. pedestrian.
134 stars 28 forks source link

模型保存时是按那个指标的? #11

Open guohaochen opened 2 years ago

guohaochen commented 2 years ago
      模型保存时是按那个指标的?`fi = fitness(np.array(results).reshape(1, -1))  # fitness_i = weighted combination of [P, R, mAP, F1]`

您好,我看过你的论文,觉得你的工作非常好。其中我注意到,你的工作中只关注了F1 score这一个指标,恰好我的工作中也使用到这个指标,所以想问问你一个问题。 yolov5中保存模型会计算一个最优值,计算的是map0.5和map的加权和 fi = fitness(np.array(results).reshape(1, -1) w = [0.0, 0.0, 0.1, 0.9] # weights for [P, R, mAP@0.5, mAP@0.5:0.95] 我想请问下你的工作保存模型是以那个指标来进行保存的,我看到你的注释中写道 # weighted combination of [P, R, mAP, F1] 请问是map和F1的加权和吗,权重比例是多少,是0.1 map+0.9 F1吗? 还是只取最大的F1 score?

zhangzhengde0225 commented 2 years ago

保存模型与Yolov5的方式一致,即0.1mAP@0.5+0.9mAP@0.5:0.95。论文分析主要根据evaluate.py获得的F1分数进行。

guohaochen commented 1 year ago

保存模型与Yolov5的方式一致,即0.1_mAP@0.5+0.9_mAP@0.5:0.95。论文分析主要根据evaluate.py获得的F1分数进行。

您好,感谢你的回复,我还有另外一个问题想请教你,请问你在进行模型训练时有没有使用YOLOV5的COCO权重作为预训练权重?还是重头开始训练的?

zhangzhengde0225 commented 1 year ago

重头训练的

在 2022年10月21日 @.***> 写道:

@.**@.:0.95。论文分析主要根据evaluate.py获得的F1分数进行。 您好,感谢你的回复,我还有另外一个问题想请教你,请问你在进行模型训练时有没有使用YOLOV5的COCO权重作为预训练权重?还是重头开始训练的? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>