dataset about transgeo2022 HOT 9 CLOSED

You can ask the authors of CVUSA (https://mvrl.cse.wustl.edu/datasets/cvusa/) for help. I do not have the right to release other people's datasets, but you can let me know about the issues. You are also free to use the VIGOR dataset (https://github.com/Jeff-Zilence/VIGOR). Btw, the accuracy on CVUSA is almost 100% and it might be better to work on other datasets.

from transgeo2022.

Thank you for your reply. I have another question. I only have one card. During the training process, there was a memory overflow and the process was killed. My server memory is 16g, not the graphics memory overflow. Is it due to Pytorch code or larger hardware required

from transgeo2022.

It should be the "accuracy" function in “train.py” that uses large memory. You can check the function and use a smaller size of queries. Maybe you can fit it in.

from transgeo2022.

I will close the issue if there is no further question.

from transgeo2022.

Thank you for your reply. It has been very helpful to me。The worker thread for data loading can only be set to 0, otherwise it will cause a crash. Is there any solution

from transgeo2022.

We use the standard distributed dataloader from Pytorch. You may try using the normal data loader to replace the distributed one.

from transgeo2022.

When training deep learning network models, the large-scale image datasets used (such as tens of thousands, hundreds of thousands, millions, etc.) contain damaged or damaged images. After a few hours of first iteration, the dataset loading problem suddenly interrupted, resulting in high time costs. How did you solve this problem

from transgeo2022.

We did not encounter this issue, but you can write an exception catcher in the data loader.

from transgeo2022.

I will close the issue if there is no further question.

from transgeo2022.