search

valeoai / xmuda

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

Current code does not support pseudo label generation for target training set #7

Closed chester256 closed 4 years ago

chester256 commented 4 years ago

When set the mode as "train", the "output_orig" will be false, while validate.py requires output_orig. https://github.com/valeoai/xmuda/blob/8b39f8f77f048cd4c086323dd89fbf9710335d35/xmuda/test.py#L75 https://github.com/valeoai/xmuda/blob/8b39f8f77f048cd4c086323dd89fbf9710335d35/xmuda/data/build.py#L44

maxjaritz commented 4 years ago

The train or test mode is set independently from the dataset. You can have mode='test' and validate on the target training set.

In the README.md, I give the following example:`

python xmuda/test.py --cfg=configs/nuscenes/usa_singapore/xmuda.yaml --pselab @/model_2d_100000.pth @/model_3d_100000.pth DATASET_TARGET.TEST "('train_singapore',)"

As you use the test script test.py, the dataloader is in test mode (output_orig = true), but you load the train_singapore (target) set and generate the pseudo labels.