[英]Loss of Object detection model using ssd_mobilenet_v2_quantized_300x300_coco increases after every 10k-12k steps
I'm re-training ssd_mobilenet_v2_quantized_300x300_coco
object detection model on a custom dataset. 我正在自定义数据集上训练
ssd_mobilenet_v2_quantized_300x300_coco
对象检测模型。 The dataset consists of approx 2.6k images and 19 classes. 该数据集包含约2.6k图像和19个类别。 After the training step reaches 10k-12k the loss graph starts increasing.
训练步骤达到10k-12k后,损耗图开始增加。 This happens even if I change my model to
ssd_mobilenet_v2_coco
and at the same step range. 即使我将模型更改为
ssd_mobilenet_v2_coco
并且在相同的步距范围内, ssd_mobilenet_v2_coco
发生这种情况。 I couldn't find anything that is related to this behaviour in the config file. 我在配置文件中找不到与此行为相关的任何内容。 Also this disappers when using
faster_rcnn
models. 当使用
faster_rcnn
模型时, faster_rcnn
消失。 When the issue arises the mAP becomes almost constant. 当问题出现时,mAP几乎变得恒定。 Also tha accuracy doesn't go beyond 50%.
准确度也不会超过50%。 Can anyone explain this behaviour ?
谁能解释这种行为?
Sample Dataset: 样本数据集:
Loss Graph 损失图
a) ssd_mobilenet_v2_quantized_300x300_coco a)ssd_mobilenet_v2_quantized_300x300_coco
b) ssd_mobilenet_v2_coco b)ssd_mobilenet_v2_coco
Config File: a) ssd_mobilenet_v2_quantized_300x300_coco 配置文件:a)ssd_mobilenet_v2_quantized_300x300_coco
https://pastebin.com/BBwqEruK https://pastebin.com/BBwqEruK
b) ssd_mobilenet_v2_coco b)ssd_mobilenet_v2_coco
What about your training loss? 那你的训练损失呢? Notice that
total_loss
is the validation loss here. 请注意,
total_loss
是此处的验证损失。
If your training loss is decreasing while the validation loss is increasing, this is clearly a sign of overfitting, you may use regularization loss during training by adding the following in the config file, in part train_config
如果您的训练损失在减少,而验证损失在增加,这显然是过度拟合的迹象,您可以在训练期间通过在配置文件中的
train_config
部分中添加以下内容来使用正则化损失
add_regularization_loss: true
just as batch_size: 24
就像
batch_size: 24
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.