search

valeoai / xmuda

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

Training Time #1

Closed AmingWu closed 4 years ago

AmingWu commented 4 years ago

It's very interesting. How long does the model need to train?

maxjaritz commented 4 years ago

Training 2D and 3D at the same time (standard setup), the training times are approx. as follows for the A2D2/Sem.KITTI scenario: baseline 30h xMUDA 50h xMUDA_PL 50h

For the NuScenes splits, the training time is faster (~26h for baseline and ~36h for xMUDA/xMUDA_PL)

Note that training can vary a lot depending on available CPU (SparseConvNet uses a lot of CPU resources).

Hope it helps.