autonomousvision/unimatch

Meaning of the pretrained model

Steven-Xiong opened this issue · 1 comments

Hello, thanks for your excellent work! I am just a little confused about the meaning of your pre-trained model. For instance, what is the meaning of the pretrained model 'gmstereo-scale2-regrefine3-resumeflowthings-sceneflow-f724fee6.pth'? Is it trained on both the flowthings dataset and sceneflow dataset, or trained on flowthings dataset and finetune on sceneflow dataset? How about the 'gmstereo-scale2-resumeflowthings-sceneflow-48020649.pth' pretrained model? Thank you.

Hi, thank you for your interest in our work.

As described in https://github.com/autonomousvision/unimatch/blob/master/MODEL_ZOO.md:

  • The models are named as model-dataset.
  • The *-resumeflowthings-* denotes that the models are trained with GMFlow model as initialization, where GMFlow is trained on Chairs and Things dataset for optical flow task.

Thus the model gmstereo-scale2-regrefine3-resumeflowthings-sceneflow-f724fee6.pth means using things pretrained GMFlow model as initilization, and then train for the stereo task on sceneflow.

Such naming applies to gmstereo-scale2-resumeflowthings-sceneflow-48020649.pth similarly.