有办法替换Visomaster、rope自带的XSeg模型吗？

wtxx8888

sinksky 发表于 2025-7-2 21:46
那大佬简单告知一下如何修改代码也行

只能告诉你，想法可以实现。
代码，不想花钱就请自学python(本人就是不满DFL，现自学的python，然后改)。。。或付费请人。

Cxsmo

On my discord server I have a patch for DeepAscension that allows you to do xseg to onnx in rope format

july

https://github.com/ykk648/XSegNet2onnx
用这个转

sinksky

july 发表于 2025-7-3 08:27
https://github.com/ykk648/XSegNet2onnx
用这个转

# usage
## generate you own onnx file
1. put DeepFaceLab XSegNet weights to weights folder, such as `weights/XSeg_256.npy`.
3. Then run `python test_seg.py` to generate tensorflow SavedModel format checkpoint file to `saved_model` directory.
5. convert model to onnx file, `python -m tf2onnx.convert --saved-model ./saved_model/ --output xseg.onnx  --tag serve`.
6. (optinal) install onnxsim `pip install onnxsim` and run `onnxsim ./xseg.onnx ./xseg_sim.onnx`.
7. `python onx2torch.py`

第三步卡住了，已经将XSeg_256.npy放入weights文件夹，但不知道怎么运行py，搜索了下用cmd执行python test_seg.py，除了弹个记事本出来，没看到生成文件