huangzongmou / yolov8_Distillation

GNU Affero General Public License v3.0
68 stars 13 forks source link

尝试多版模型,均没有获得蒸馏提升 #10

Open zzzz737 opened 4 months ago

zzzz737 commented 4 months ago

尝试教师模型:s, m, l;学生模型:n,蒸馏方式 cwd 或者 mgd, 在coco上训练,准确率都没有提升,反而有略微下降,不解!

huangzongmou commented 3 months ago

你要对比你自己训练的n模型。不是官网给的指标。

---原始邮件--- 发件人: @.> 发送时间: 2024年3月4日(周一) 下午5:29 收件人: @.>; 抄送: @.***>; 主题: [huangzongmou/yolov8_Distillation] 尝试多版模型,均没有获得蒸馏提升 (Issue #10)

尝试教师模型:s, m, l;学生模型:n,蒸馏方式 cwd 或者 mgd, 在coco上训练,准确率都没有提升,反而有略微下降,不解!

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>

minushuang commented 2 months ago

尝试教师模型:s, m, l;学生模型:n,蒸馏方式 cwd 或者 mgd, 在coco上训练,准确率都没有提升,反而有略微下降,不解!

对比自己训练的n模型,与蒸馏之后的n模型,指标有提升吗?