search

zhang-tao-whu / DVIS

DVIS: Decoupled Video Instance Segmentation Framework
MIT License
127 stars 6 forks source link

Training parameters #12

Closed nanle771 closed 1 year ago

nanle771 commented 1 year ago

How can I replicate similar results on only one GPU? What parameters need to be modified? Can I display only the segmentation mask (without categories and scores) when running the demo.py for visualization? Thank you!

zhang-tao-whu commented 1 year ago

Perhaps you could decrease the learning rate and batch size while increasing the number of iterations accordingly. For example, BASE_LR: 0.0000125, IMS_PER_BATCH: 1, MAX_ITER: 160000, STEPS: (112000,).

zhang-tao-whu commented 1 year ago

If you want to hide the class and score, please modify the draw_instance_predictions function on line 50 of visualizer.py.