FCHD-Fully-Convolutional-Head-Detector
FCHD-Fully-Convolutional-Head-Detector copied to clipboard
aditya-vora
为什么做了pytorch的数据增强后效果反而变差了
为了提高模型的精度，我想到了如下三个办法1：做增量训练，2：改用Adam优化器，暂时还没有成功，具体报错是数据格式的问题，我稍后再尝试一下,3:做数据增强，可是模型评估的系数一直很低，所以想请教一下有什么办法提议提高模型的精确度，还有就是判断标准就就仅仅看每一个epochs后打印的那个准确值吗??在一定的epoch后选择一个最好的模型再继续往下训练会有效果吗？？个人感觉会过拟合，而且评估系数只会越来越低，这又该如何解释呢。
In order to improve the accuracy of the model, I thought of the following three methods: 1 to do incremental training, 2: to use the Adam optimizer, for the time being has not been successful, the specific error is the data format problem, I will try again later, 3: Data enhancement, but the coefficient of model evaluation has been very low, so I would like to ask if there is any way to propose to improve the accuracy of the model, and the criteria for judging just to see the exact value printed after each epochs? After selecting the best model after epoch, will it be effective to continue training? ? Personal feelings will be over-fitting, and the evaluation coefficient will only get lower and lower, so how to explain it.
