search

Megvii-BaseDetection / YOLOX

YOLOX is a high-performance anchor-free YOLO, exceeding yolov3~v5 with MegEngine, ONNX, TensorRT, ncnn, and OpenVINO supported. Documentation: https://yolox.readthedocs.io/
Apache License 2.0
9.47k stars 2.21k forks source link

High training loss #284

Open HUST-WangZhimin opened 3 years ago

HUST-WangZhimin commented 3 years ago

作者你好！我用你们的yolox训练自己的数据集时发现total_loss一直震荡下不去，我使用的数据集一共有3000张图片，训练集验证集的划分比例为8：2。请问需要进行哪些改动才可以让total_loss降下去呢？谢谢！ 1 2

scuizhibin commented 3 years ago

我的一直在4.4左右徘徊，我使用的是voc数据集和。

Joker316701882 commented 3 years ago

@HUST-WangZhimin How is your evaluation mAP?

scuizhibin commented 3 years ago

map5095  57.6  map50  83。这个怎么样？voc数据集和上训练的。

---原始邮件--- 发件人: "Ge @.> 发送时间: 2021年7月31日(周六) 下午5:14 收件人: @.>; 抄送: @.**@.>; 主题: Re: [Megvii-BaseDetection/YOLOX] 训练损失一直下不去 (#284)

@HUST-WangZhimin How is your evaluation mAP?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

Joker316701882 commented 3 years ago

@scuizhibin This is normal performance.

HUST-WangZhimin commented 3 years ago

@HUST-WangZhimin How is your evaluation mAP?

The evaluation mAP is around 0.83，I trained 100 epochs

scuizhibin commented 3 years ago

map  0.83

---Original--- From: @.> Date: Mon, Aug 2, 2021 17:24 PM To: @.>; Cc: @.**@.>; Subject: Re: [Megvii-BaseDetection/YOLOX] High training loss (#284)

@HUST-WangZhimin How is your evaluation mAP?

The evaluation mAP is around 0.83，I trained 100 epochs

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

scuizhibin commented 3 years ago

我加入了注意力结果也才有0.84，最小的模型。

---Original--- From: @.> Date: Mon, Aug 2, 2021 17:24 PM To: @.>; Cc: @.**@.>; Subject: Re: [Megvii-BaseDetection/YOLOX] High training loss (#284)

@HUST-WangZhimin How is your evaluation mAP?

The evaluation mAP is around 0.83，I trained 100 epochs

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

scuizhibin commented 3 years ago

0.84map在voc数据上，我加入了注意力机制，提升的不明显啊

---原始邮件--- 发件人: "Ge @.> 发送时间: 2021年7月31日(周六) 下午5:21 收件人: @.>; 抄送: @.**@.>; 主题: Re: [Megvii-BaseDetection/YOLOX] 训练损失一直下不去 (#284)

@scuizhibin This is normal performance.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

charles-str commented 3 years ago

我的一直在4.4左右徘徊，我使用的是voc数据集和。

你好，你训练的是yolox-tiny的版本吗？交流下？

charles-str commented 3 years ago

作者你好！我用你们的yolox训练自己的数据集时发现total_loss一直震荡下不去，我使用的数据集一共有3000张图片，训练集验证集的划分比例为8：2。请问需要进行哪些改动才可以让total_loss降下去呢？谢谢！ 1 2

您训练的是yolox-tiny吗？交流下

scuizhibin commented 3 years ago

不是 tiny版本

---Original--- From: @.> Date: Wed, Oct 20, 2021 15:42 PM To: @.>; Cc: @.**@.>; Subject: Re: [Megvii-BaseDetection/YOLOX] High training loss (#284)

我的一直在4.4左右徘徊，我使用的是voc数据集和。

你好，你训练的是yolox-tiny的版本吗？交流下？

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.