简体繁体 English

SSD Resnet 50 FPN Loss 函数说明

[英]SSD Resnet 50 FPN Loss function clarification

原文 2020-02-25 07:23:09 7 1 tensorflow/ machine-learning/ computer-vision/ loss-function/ object-detection-api

I am using tensorflow object detection api on my dataset.我在我的数据集上使用 tensorflow 对象检测 api。 I am using ssd-resnet50-fpn model.我正在使用 ssd-resnet50-fpn 模型。 While training, I see that classification loss and localization loss has converged but the total loss is still decreasing.在训练时，我看到分类损失和定位损失已经收敛，但总损失仍在下降。 Also total loss is not coming out to be the sum of classification loss and localization los.总损失也不是分类损失和定位损失的总和。 Any ideas on why this is happening.关于为什么会发生这种情况的任何想法。 I am using train.py in object_detection/legacy/ folder to train on my dataset.我在 object_detection/legacy/ 文件夹中使用 train.py 来训练我的数据集。 Attached image for the same.附上相同的图像。

1 个解决方案

总损失是应用于可训练变量的分类损失、定位损失和 L2 损失的总和，并由“weight_decay”加权

使用 ResNet50，验证准确度和损失不会改变 - With ResNet50 the validation accuracy and loss is not changing

如何将ssd_resnet_50 tensorflow checkpoint转换为.tflite？ - How to convert ssd_resnet_50 tensorflow checkpoint to .tflite?

Tensorflow Object Detection API - Transfer Learning on "CenterNet Resnet50 V1 FPN 512x512" model error - Tensorflow Object Detection API - Transfer Learning on "CenterNet Resnet50 V1 FPN 512x512" model error

如何在 Keras 或 Tf 2 中为基于 ResNet50 的连体网络应用 Triplet Loss - How to apply Triplet Loss for a ResNet50 based Siamese Network in Keras or Tf 2

训练 ResNet50 时高 val_loss 和低 val_accuracy model - High val_loss and low val_accuracy when training ResNet50 model

ResNet50在喀拉拉邦进行转移学习时，损失总是归功于南 - Loss always go to nan when transfer learning by ResNet50 in keras

是否可以在 Android 上运行 ssd_mobilenet_v1_fpn 进行推理？ - Is it possible to run ssd_mobilenet_v1_fpn on Android for inference?

无法在 android 上部署 ssd mobilenet v1 fpn - Unable to deploy ssd mobilenet v1 fpn on android

SSD-shufflenet-V2-FPN比Mobilenet V2慢 - SSD-shufflenet-V2-FPN is Slower than Mobilenet V2

验证损失在 Resnet 中没有改变 - Validation loss not changing in Resnet

暂无

暂无

声明:本站的技术帖子网页，遵循CC BY-SA 4.0协议，如果您需要转载，请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.

相关问题 使用 ResNet50，验证准确度和损失不会改变 - With ResNet50 the validation accuracy and loss is not changing 如何将ssd_resnet_50 tensorflow checkpoint转换为.tflite？ - How to convert ssd_resnet_50 tensorflow checkpoint to .tflite? Tensorflow Object Detection API - Transfer Learning on "CenterNet Resnet50 V1 FPN 512x512" model error - Tensorflow Object Detection API - Transfer Learning on "CenterNet Resnet50 V1 FPN 512x512" model error 如何在 Keras 或 Tf 2 中为基于 ResNet50 的连体网络应用 Triplet Loss - How to apply Triplet Loss for a ResNet50 based Siamese Network in Keras or Tf 2 训练 ResNet50 时高 val_loss 和低 val_accuracy model - High val_loss and low val_accuracy when training ResNet50 model ResNet50在喀拉拉邦进行转移学习时，损失总是归功于南 - Loss always go to nan when transfer learning by ResNet50 in keras 是否可以在 Android 上运行 ssd_mobilenet_v1_fpn 进行推理？ - Is it possible to run ssd_mobilenet_v1_fpn on Android for inference? 无法在 android 上部署 ssd mobilenet v1 fpn - Unable to deploy ssd mobilenet v1 fpn on android SSD-shufflenet-V2-FPN比Mobilenet V2慢 - SSD-shufflenet-V2-FPN is Slower than Mobilenet V2 验证损失在 Resnet 中没有改变 - Validation loss not changing in Resnet

相关标签

粤ICP备18138465号 © 2020-2024 STACKOOM.COM