• I have trained and tested the program successfully, and the validation is very good (always 99.5%, even if with only 100 steps), but when I run the label_image.py, the result is very bad: almost all ...
• 训练时validation accuracy和train accuracy之间没有差距，本义为这是一个还不错的曲线，但是今天讨论时有人评论说这种情况说明网络参数不足，因为在参数充足的情况下多多少少会有过拟和，所以正常情况下train ...
训练时validation accuracy和train accuracy之间没有差距，本义为这是一个还不错的曲线，但是今天讨论时有人评论说这种情况说明网络参数不足，因为在参数充足的情况下多多少少会有过拟和，所以正常情况下train accuracy会高于validiation accuracy，似乎很有道理！


展开全文
• <div><p>In running the sample code, I found that the training accuracy is much lower than the validation accuracy, which is different from training GraphSAGE on reddit in their repo. Is this normal. ...
• When I use your uploaded <code>VOC12_scenes_20000.pth</code> model to evaluate the single scale model on 1449 validation data, I get 71.8% which is much lower than reported 74.0%. Moreover, when I try...
• <div><p>In the following section: <code>Open Standards > Benefits > Interoperability > Cloud computing</code></p> <p>I don't feel that the following statement is accurate. ...
• t freeze any layer of it, after 40 epochs iteration, it still get too bad validation accuracy in my validation datasets, just as the following picture: so, I just want to know whether is's nornal...
• 使用davidsandberg大侠的facenet 代码时， https://github.com/davidsandberg/facenet , 对于训练结果评价，提示类似如下:Accuracy: 0.99650+-0.00252Validation rate: 0.98367+-0.00948 @ FAR=0.00100 可以看到有...


使用davidsandberg大侠的facenet  代码时， https://github.com/davidsandberg/facenet , 对于训练结果评价，提示类似如下:Accuracy: 0.99650+-0.00252Validation rate: 0.98367+-0.00948 @ FAR=0.00100
可以看到有两个评价方法，accracy 和 validation rate。 我对 accracy 和 val rate的计算方式是明白的，但中间的逻辑（为什么要这样算） 有疑问，很长时间都没弄明白..
首先  计算方式如下：
accuracy              =（判断正确的 同一个人+ 判断正确的 不同的人） / 所有数据
validation rate     =  判断正确的同一个人的数量 / 所有的同一个人 的数量        (这时有个数字 代码写死为0.001，后面细说)
但为什么要有两个指标评价? accuracy不就够了么? 而且貌似val rate比accuracy 更重要，更能反映训练结果的好坏。

开始时，我的训练的结果是acc 90%, val rate 10%， 后来提升到了 acc 98%, val rate 60%

我的疑问/思路，举个例子来说 当acc 90%, val rate 10%时:

设所有数据是100份，  现在我的accuraccy为90%，则 （判断正确 同一个人+ 判断正确 不同的人 ）总和为90人，
假设同一人的情况 实际数量为50：10%val则 5个判断对，则判断正确不同的人需要90-5=85个，  50+85>100， 不可能
假设同一人的情况 实际数量为80：10%val则 8个判断对，则判断正确不同的人需要90-8=72个，  80+72>100， 不可能
假设同一人的情况 实际数量为10：10%val则 1个判断对，则判断正确不同的人需要90-1=89个，  10+89约等于100，有可能
而最后一种就是不平衡数据集，   而我的测试集是比较平衡的，不会100组数据中只有10个人左右是相同的 。 问题到底在哪呢?
先看/理解好 知乎这条问题:

https://www.zhihu.com/question/30750849

现在看垂直的这条分界线 也就是我们最后要设定的判断阈值，把交叉的部分分成了假正 假负 两块区域。

现在把 误报率 也就是假正 ，变为很小的0.001，就是要把蓝色的线 移到靠近右边的虚线。(把假正降到很小很小的0.001)
而这时，validtateion rate不高，就说明我们训练的结果 还不够分得开，交叉的地方太大了：假正假负都太多， 我们需要 假正假负都尽量小

如果我们的是模型训练得好（类似下图，交叉的地方很小），误报率 即使设为很低的0.001， val rate也可以高。下图是我们的训练目标， 上图是我们还没训练得足够好的情况。

所以，与不平衡数据集没关系，是我们的网络训练得不够。
那为什么模型训练得不好，accuracy仍可达90%或更高以上呢? 我想像中的，只有达到上图的效果了，accuracy才会达到99%， 实际下图这种情况，也可以达到accuracy99%, validation rate 60%。

另外红线和蓝线，分别就是accuracy和validation rate 的门限值设置方法。也就是为什么有两个评价标准，且val rate比accuracy 更重要，更能反映训练结果的好坏。


展开全文
• In detail, after some epoches of training, resnet50 and se-resnext dygraph training accuracy improve as expected, but validation accuracy is much lower than training accuracy. <p>We do not understand...
• <p>When running train.py I get good accuracy on train data but 0.000 accuracy on validation data. At the end all precision, recall and f-1 score is all 0. <p>Could you please address this problem. ...
• <p>I work pretty regularly with PyTorch and ResNet-50 and was surprised to see the ResNet-50 have only 75.02% validation accuracy. When I use the pretrained ResNet-50 using the code ...
• <div><p>训练参数为batch size=...而validation acc突然减小至不到0.1且随着训练validation acc会继续下降。请问我的训练参数该如何调整？</p><p>该提问来源于开源项目：MVIG-SJTU/AlphaPose</p></div>
• <div><p>Using the code below, we see that accuracy improves on training data but not on validation. Here I used a subset (891 training images. 110 validation) of the Carvana Kaggle Image Masking ...
• However, the experiments on ImageNet with AlexNet achieved about 3% higher accuracy on the ImageNet validation set than the published results. <p><code>supervised</code>: 59.48 (59.70 in the paper) ...
• <div><p>Hello , I am trying to train the network on NVIDIA Titan RTX 24 GB GPU, I find that both val_loss and val_acc are increasing, while train_loss is reducing and train_acc is increasing as ...
• <div><p>Caffe has support wherein if you provide 2 top output blobs for the final accuracy layer, it displays a per-class accuracy as well during the testing phase. <p>Is it possible to get something ...
• 知乎：深度学习为什么会出现validation accuracy大于train accuracy的现象？ 概括他们的答案： validation的样本数量一般远小于training的 val的时候是用已经训练了一个epoch的model进行测试的(经过大量的训练学习...
知乎：深度学习为什么会出现validation accuracy大于train accuracy的现象？
概括他们的答案：

validation的样本数量一般远小于training的
val的时候是用已经训练了一个epoch的model进行测试的(经过大量的训练学习到东西了)
data augmentation把训练集变得丰富，制造数据的多样性和学习的困难来让network更robust（比如旋转，随机crop，scale），但是val和test的时候一般是不对数据进行data augmentation的
各种正则化，dropout在训练集上使用，却不会在验证集上使用，导致训练集的 loss 偏大


展开全文
• For training from scratch, fine-tuning and parallel adapter training, the validation accuracy for the aircraft data set I get is 38%, 50%, 44%, however the published test accuracy are 57.10%, 60.34%,...
• presenting the DNN accuracy. We can also predict the sequence using beam search and get a CER report in the stage "Decoding". However, A high classification accuracy in DNN but a low ASR ...
• The validation mIoU of VOC_scenes_20000.pth is 71.1. Is this degradation due to randomness? Could you give me any comments? Here is the parameter I used: <p>BATCH_SIZE = 10 DATA_DIRECTORY = &...
• ve managed to get to over 80% training accuracy while the validation accuracy stays below 20% and is unfortunately seems to be random and nonmonotonic. <p>I think there's an issue with how you are...
• Am I misunderstanding the graphs reporting validation accuracy on the training job page? (I've attached a sample graph.) <p><img alt="reported+accuracy" src=...
• <p>The mIoU results that I achieve are at best around 62% mIoU on the Cityscapes validation set for training from scratch. Now I am wondering if I am missing something during training since your ...
• <div><p>I ran the training script ...Accuracy : 89.38 Error : 10.62 <p>Which seems very low compared to the 94.74% in the readme and paper.</p><p>该提问来源于开源项目：titu1994/DenseNet</p></div>
• <p>However, using the <code>lorra_best.pth</code> checkpoint achieves only <code>15.60%</code> vqa accuracy. All the code is the same as master, I am using the below command to run inference: <pre>...
• <div><p>I have added a model which reaches 83% accuracy on validation set. By default the LSTM model will be chosen. There is a command line parameter called model_choice which switches to the second ...
• I have trained Resnet-20 on cifar-10 using KFAC several times and all the experiments reveals that although KFAC converges faster, but the final validation accuracy is not as good as momentum-...
• <div><p>As with #5, the validation accuracy for the large model is also well below the stated. I was curious because the stated result, beating the official, with 1.4m less parameters would be ...
• ] - 252s 492ms/step - loss: 0.1199 - accuracy: 0.9728 - val_loss: 7.3933e-05 - val_accuracy: 1.0000 saved /content/checkpoints.model.0 Finished Epoch 0 <p>I have high accuracies but bad prediction. I ...
• when I try to apply the pre-train model(resnext-101-64f-kinetics.pth)on the validation set from Kinetic dataset, the accuracy turns out to be very low(much like random prediction). I have check the ...

...