how to infer single image?

[shining-love]

Hello, may I ask how to perform inference on a single image without a ground-truth mask? This needs to support both the original trained PT model and inference with the exported ONNX model for single-image prediction. There are also concerns regarding precision alignment between the two inference approaches. Could the author please provide some inference examples based on this project?

[blaz-r]

We do not have a publicly ready-to-use inference code. You can modify the model's forward to do that, and I believe onnx should export just fine with some minor issues discussed in #10

[shining-love]

We do not have a publicly ready-to-use inference code. You can modify the model's forward to do that, and I believe onnx should export just fine with some minor issues discussed in #10

My current scenario is that I only have images of good products and cannot collect defective samples. How can I train using only the good product images? How should the dataset be organized? What changes need to be made to the code? Please help me. Thank you.

[shining-love]

We do not have a publicly ready-to-use inference code. You can modify the model's forward to do that, and I believe onnx should export just fine with some minor issues discussed in #10

In actual production, we often cannot obtain defective products and can only obtain some good products. In this case, when training only with good products, how should I train using your model code? Please give me some guidance, thank you.