Anyone has tried to train mobilenet1.0 with MNIST dataset under the directory /example/gluon, what’s the state-of-art test accuracy with mobilenet1.0, my experimental results is around 91%. Any reference has mentioned about the test accuracy with mobilenet1.0? I cannot find such papers to compare my experimental results. Help needed!!
You can check the accuracies of state of the art models that are part of the Gluon’s model zoo here: https://gluon-cv.mxnet.io/model_zoo/classification.html
However, the models listed there are trained on ImageNet and not MNIST.
Thanks. I have checked the accuracy on ImageNet while I want to know the accuracy on MNIST to verify my experimental results. Besides, the accuracy of training ResNer18_v2 with CIFAR-10 is also strange (around 81%). The following is my script: python image_classification.py --model resnet18_v2 --dataset cifar10 --gpus 0,1,2,3 --batch-size 32 --mode symbolic --lr 0.1 --epochs 200 --lr-step-epochs 80,120. I tried this on mxnet 1.3.0, 1.3.1, 0.12.1 and I do not know why this happens.