long123524/bsinet-torch

JAG: Delineation of agricultural fields using multi-task BsiNet from high-resolution satellite images

bsinet-torch's Issues

Comparison of models in the research paper

This question is related to the comparison of models in the research paper - Delineation of agricultural fields using multi-task BsiNet from high-resolution satellite images. You have compared the performance of BsiNet with three other models (UNet, Psi-Net, and ResUNet-a D7). The original implementation of ResUNet-a D7 is in mxnet DL framework. Did you translate it into PyTorch or relied on the original implementation in mxnet?

作者您好，请问您最近有计算出或者找到ST-UNet的GFlops吗？

我无法计算出ST-UNet对应文章的GFlops，碰巧看到你也遇到了同样的问题并且向作者发起了问题，由于最近急需比对所以想请问下你有找到吗？万分感谢您！

test.py

你好，我想问下，您最后有预测出distance嘛？我看到test.py代码最后只是输出了一个？还将其作为为辅助任务，最后只是输出了mask?我在最后预测修改了代码，但是输出的距离预测图就是一直不太对。

Where is the pretrain model?

I can't find the model --model_file ./model/150.pt in Evaluate stage. Can you commit it, please.

pretrained weight

Could you please upload the pretrained weight to google drive
Thank you

Load pretrained model

Hello
When I load pretrained model, i got an issue , as you can see in attached file.

So i modified test.py as like below

so i could fix tha issue,but i got different inference result whenever I run inference with that model.
How can i fix it?

作者您好，请问GOC, GUC, and GTC这些评价指标是基于矢量计算的吗？如果可以的话，能不能开源一下这部分的代码，十分感谢！

Tensor size mismatch during prediction

I received tensor mismatch error when making a prediction.

inputs = inputs.to(device = device)
outputs1, outputs2, outputs3 = model(inputs)

数据集什么时候开源？

您使用了很多GF2和GF1号数据，是否可以考虑将地理坐标脱敏后开源矢量地块标注和高分影像

请问有训练好的权重文件吗?

请问有训练好的权重文件吗?最近复现您的模型,预测结果出现错误,想用您的权重文件试一下测试结果,看看是否与模型训练有关
如果您有训练好的权重文件,麻烦您发我邮箱[email protected],感谢!

About pretrained weights

I use the pretrained weight on a Xinjiang GF-2 image，but got the The following error：

Exception has occurred: RuntimeError
Error(s) in loading state_dict for DataParallel:
Missing key(s) in state_dict: "module.down.0.l1.conv.bias", "module.down.0.l2.conv.bias", "module.down.1.l1.conv.bias", "module.down.1.l2.conv.bias", "module.down.2.l1.conv.bias", "module.down.2.l2.conv.bias", "module.down.3.l1.conv.bias", "module.down.3.l2.conv.bias", "module.down.4.l1.conv.bias", "module.down.4.l2.conv.bias", "module.up.0.l1.conv.bias", "module.up.0.l2.conv.bias", "module.up.1.l1.conv.bias", "module.up.1.l2.conv.bias", "module.up.2.l1.conv.bias", "module.up.2.l2.conv.bias", "module.up.3.l1.conv.bias", "module.up.3.l2.conv.bias".

Conversion of the segmentation results (raster format) into field boundaries (vector format)

How were the segmentation results (raster format) converted into field boundaries (vector format)?