search

Alpha-VL / ConvMAE

ConvMAE: Masked Convolution Meets Masked Autoencoders
MIT License
477 stars 41 forks source link

how many gpus and which type of gpu are needed for pretraining, down-stream task finetuning? #4

Closed wenmengzhou closed 2 years ago

wenmengzhou commented 2 years ago

specifically, gpu type, gpu number and gpu training time for pretraining, detection training and segmentation training

gaopengpjlab commented 2 years ago

Pretraining is done with 8 A6000 with 48G GPU memory Detection is done with 8 A100 with 80G GPU memory which tooks around 1 day and 5 hours Segmentation is done with 8 V100 with 32G GPU memory

wenmengzhou commented 2 years ago

Pretraining is done with 8 A6000 with 48G GPU memory Detection is done with 8 A100 with 80G GPU memory which tooks around 1 day and 5 hours Segmentation is done with 8 V100 with 32G GPU memory

thanks

gaopengpjlab commented 2 years ago

Please check FastConvMAE for a fast implementation of pretraining and also benchmarking different MIM paradigm.

wenmengzhou commented 2 years ago

Please check FastConvMAE for a fast implementation of pretraining and also benchmarking different MIM paradigm.

pls provide a link？ Google failed to find it

gaopengpjlab commented 2 years ago

https://github.com/Alpha-VL/FastConvMAE. Code (one or two days) and paper (one week) will be released soon.