简体   繁体   English

创建 ML 始终“在 10 次迭代中尽早收敛”

[英]Create ML always "converged early at 10 iterations"

I'm trying to train object detection model with dataset about 100k images, for recognise 7-segment digits (about 10k 100x100 grayscale images per digit) and Create ML stops training with the same result "converged early at 10 iterations", regardless of the accuracy and quality of the data (I started with a dataset of 20k images).我正在尝试使用大约 100k 图像的数据集训练 object 检测 model,以识别 7 段数字(每个数字约 10k 100x100 灰度图像)并创建 ML 停止训练,结果相同“在 10 次迭代中早期收敛”数据的准确性和质量(我从 20k 图像的数据集开始)。

My last try:我的最后一次尝试:

  • Training Accuracy 68,4%训练准确率 68.4%
  • Validation Accuracy 68,2%验证准确度 68.2%

Training process培训过程

The answer is - don't try to use Create ML for training models)) I ran the same training on google colab using this turicreate template and got 94% accuracy without interruptions at 50 iterations.答案是——不要尝试使用 Create ML 来训练模型))我使用这个 turicreate 模板在 google colab 上进行了相同的训练,并且在 50 次迭代中没有中断地获得了 94% 的准确率。

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.

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