Karel

Hi @liuzhihui2046, A1. Exactly. Before multiplying the y_ij, the omega contains not only the foreground but also background areas. Thus, we exclude the bg area using y_ij (GTs). As a...

We have released the [pre-trained TRACER](https://github.com/Karel911/TRACER/releases/tag/v1.0) and you can reproduce the results using `test` action. If you need more detailed information for the inference, please refer to the [Readme.md, section...

Hi, Actually, we did not compare TRACER with the U2Net in our research. Based on the reported results of [U2Net](https://arxiv.org/pdf/2005.09007.pdf) on benchmark datasets, our TRACER outperformed U2Net for the five...

The multi-class classification you mean is such a semantic segmentation task?

Hi @DeepHM, Sorry for the late reply. Does your network return the shape of X \in R^{B, n_classes, H, W}?

Sorry for the late reply. Did you try to decrease learning rate?

Hi, sorry for late rely. I guess there are some values having < -1 or > 1 in your label. please check your gt (mask) values.

Please change the num of batch sizes or num_workers. It is mainly originated from CUDA, not the codes.

Did you use a single gpu? If yes, change `multi_gpu=False` in [config](https://github.com/Karel911/TRACER/blob/main/config.py#L37). Then, try num_workers=0. When we tried it with the same device, we could not reproduce the error.

It seems there are negative or nan values in your output or labels. Please check the values.