Closed wjtan99 closed 5 years ago
Accuracy>0.87 can be achieved with default settings. Maybe, your training set is different from http://www.vision.caltech.edu/visipedia/CUB-200.html.
I am using the CUB_200_2001 data set. But you are right, I am not using the exact train_test_split.txt. Anyway, by adjusting starting lr, and batch size, I can now achieve top1 acc>0.87. Thanks any way.
Great! Let we know if anything doesn’t work :)
I am using the CUB_200_2001 data set. But you are right, I am not using the exact train_test_split.txt. Anyway, by adjusting starting lr, and batch size, I can now achieve top1 acc>0.87. Thanks any way.
Hi guys, do you achieve the same performance on CUB_200_2011 dataset?
I retrained the model with default setting. My best result is 87.30% on CUB.
i can just get 85.95% on CUB 200-2011 dataset with default hyper parameters. Is there anything wrong?
I am using the CUB_200_2001 data set. But you are right, I am not using the exact train_test_split.txt. Anyway, by adjusting starting lr, and batch size, I can now achieve top1 acc>0.87. Thanks any way.
Hello!, thanks for your sharing, I'm wondering that how did you adjust the parameters that made you achieve top1>0.87? I'm using the original CUB200 dataset split(train : test=1 : 1), train batchsize=16, train epoch=180, start_lr=0.0008, but I can only get top1=0.86, can you share some experience about your parameters adjusting?
What are the training hyper parameters to get top1 acc>0.87, like batch, lr strategy? and how many epochs do you run? I can only get 0.83 on the CUB_200_2001. I divided the data sets to 70% training, 30% validation. Please do not close the issues so fast. Thanks.