search

yzbouc / UVEB

MIT License
44 stars 3 forks source link

Man I ask how long it took to train for 150K iterations on 4 NVIDIA Tesla A40 GPUs? #10

Open MoveForward111 opened 4 months ago

MoveForward111 commented 4 months ago

The batch size is 4, and the patch size of input video frames is 512×512.

yzbouc commented 4 months ago

Thank you for your attention！Our network was trained for about a week. You can try following our instructions. But we are not sure if this is the optimal training time (1) At that time, considering the deadline for CVPR and our limited computing resources, we could only train for one week (2) There is significant similarity between video frames, but there is little research and guidance on the appropriate training epochs for video data. (3) The training duration may need to be considered based on the number of parameters in your network.