search

valeoai / xmuda

Cross-Modal Unsupervised Domain Adaptationfor 3D Semantic Segmentation
Other
194 stars 36 forks source link

Selecting highly confident labels for pseudo label generation #9

Closed weiliuxm closed 4 years ago

weiliuxm commented 4 years ago

Thank you for sharing the code.

The paper says

In details, once having optimized a model with Eq. 4, we extract pseudo-labels offline, selecting highly confident labels based on the predicted class probability.

However, I could not find how to select highly confident labels in validate.py.

maxjaritz commented 4 years ago

You are correct. In validate.py only the maximum probability and corresponding class label are saved to a .npy file. The label refinement is done when initializing the dataloader, using xmuda/data/utils/refine_pseudo_labels.py.

Best, Max

weiliuxm commented 4 years ago

Thanks for the reply.

taeyeopl commented 3 years ago

Can you explain why you chose median operation as the highly confident label threshold instead of other options (mean, some threshold, topK)?? Or have you tried these experiments??? https://github.com/valeoai/xmuda/blob/8b39f8f77f048cd4c086323dd89fbf9710335d35/xmuda/data/utils/refine_pseudo_labels.py#L17