search

mxin262 / SwinTextSpotter

Pytorch re-implementation of Paper: SwinTextSpotter: Scene Text Spotting via Better Synergy between Text Detection and Text Recognition (CVPR 2022)
275 stars 42 forks source link

CUDA out of memory (How much VRAM is needed) #99

Closed shshwtsrkr closed 1 year ago

shshwtsrkr commented 1 year ago

Hi, I was following the instructions mentioned in the README. I am finetuning the model on CTW1500 in a kaggle notebook. Every time I use the command :-

python projects/SWINTS/train_net.py \ --num-gpus 2 \ --config-file projects/SWINTS/configs/SWINTS-swin-mixtrain.yaml

I always run into the error - CUDA ran out of memory

For additional information I am using the kaggle configuration of 2 Tesla T4 (i.e. the T4x2 setup) with 14.8GB VRAM (i.e. a total of 29.6 GB VRAM)

It would be really helpful if I could get some help regarding this. Also I would be grateful if I get to know what GPU(s) and how many GPUs are suitable for the task.

mxin262 commented 1 year ago

Hi, I used 8 A6000 for training.

shshwtsrkr commented 1 year ago

Thanks for answering, this helped a lot.