【demo.py】Change the input ground truth, the estimated transformation will change

[ljjTYJR]

Hello,

When I was using 3d-match demo, I met with a strange thing:

I was testing the demo file:

【3D-Match demo】

It needs four inputs:

• The src
• The ref
• The gt (Ground Truth)
• The weights

But when I changed the gt input, the estimated transformation also changed. In my view, the output (the estimated transformation) should be the same no matter what gt I provide.

where am I going wrong? Or is it the usual thing?

Best, Shuo.

[ljjTYJR]

Found the wrong place:

If to use pre-trained data on my own test data, the self. training should be set false

[superstrong123]

Found the wrong place:

If to use pre-trained data on my own test data, the self. training should be set false

我也有类似的问题，但是这个self.training我没找到，楼主在哪个文件找到这个参数的呢0.o？

[ljjTYJR]

Found the wrong place: If to use pre-trained data on my own test data, the self. training should be set false

我也有类似的问题，但是这个self.training我没找到，楼主在哪个文件找到这个参数的呢0.o？

https://github.com/qinzheng93/GeoTransformer/blob/main/experiments/geotransformer.3dmatch.stage4.gse.k3.max.oacl.stage2.sinkhorn/model.py line 163

顺着模型的逻辑看下去就可以找到。