search

Gumpest / YOLOv5-Multibackbone-Compression

YOLOv5 Series Multi-backbone(TPH-YOLOv5, Ghostnet, ShuffleNetv2, Mobilenetv3Small, EfficientNetLite, PP-LCNet, SwinTransformer YOLO), Module(CBAM, DCN), Pruning (EagleEye, Network Slimming), Quantization (MQBench) and Deployment (TensorRT, ncnn) Compression Tool Box.
989 stars 202 forks source link

TPH-YOLOV5的训练结果 #26

Open hu874 opened 2 years ago

hu874 commented 2 years ago

您好，请问为什么我使用yolov5x-TPH训练小目标数据集的最终map只有95.5，而使用yolov5x训练同样的数据集最终map有99，请问这是什么原因呢？

Gumpest commented 2 years ago

您好，请您观察训练集的Loss是否过拟合。

hu874 commented 2 years ago

您好，训练集的loss正常，为何改良的yolov5效果不明显，能帮帮吗？

------------------ 原始邮件 ------------------ 发件人: ZhangYuan @.> 发送时间: 2021年12月2日 22:10 收件人: Gumpest/YOLOv5-Multibackbone-Compression @.> 抄送: hu874 @.>, Author @.> 主题: Re: [Gumpest/YOLOv5-Multibackbone-Compression] TPH-YOLOV5的训练结果 (Issue #26)

您好，请您观察训练集的Loss是否过拟合。

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