- Imbalanced Data classes - It could happen that a few classes have many data points and other classes have a few points. To remove this ambiguity for the model, kindly follow the link.
- Data Augmentation - You can augment the images (link). This helps to increase the accuracy of the model by giving a few more training points.
Testing accuracy of pretrained model (resnet18)
7 次查看(过去 30 天)
显示 更早的评论
I tested the pretrained resnet18 network available on MATLAB using CIFAR-100 dataset (100 class dataset) and the accuracy is very low, even though resnet18 is an object classification network.
How can I get the actual accuracy? because I saw in some papers using the same CIFAR dataset they got around 70% accuracy.
0 个评论
回答(1 个)
Gaurav Garg
2021-1-28
Hi Nour,
I hope you are using transfer learning to train the network 'resnet-18'. You can look at example here for more guidance on transfer learning, though the given example is for AlexNet.
If the model still doesn't give good accuracy, it can be because of many reasons, a few of which have been listed here:
0 个评论
另请参阅
类别
在 Help Center 和 File Exchange 中查找有关 Image Data Workflows 的更多信息
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!