flyinglynx / Bilinear-Matching-Network

Official implementation for CVPR 2022 paper "Represent, Compare, and Learn: A Similarity-Aware Framework for Class-Agnostic Counting".
MIT License
69 stars 9 forks source link

How much RAM is required for training? #2

Closed zyweven closed 2 years ago

zyweven commented 2 years ago

Now I have 24G RAM and 12G GPU memory. But when I train, I always get this in epoch 1

RuntimeError: DataLoader worker (pid 15666) is killed by signal: Killed. 

This seems to be the reason for not having enough RAM . How much RAM is required during training? Thank you very much

flyinglynx commented 2 years ago

Hi! Actually, the requirement for RAM is relatively high (around 30G), because we preload the whole dataset into the memory in our code. But this is avoidable at the cost of some additional training time (reading data from disk). Please change default value for preload argument in line 64 of FSC147_dataset.py from True to False. We will modify the code so that the preload choice for dataset can be modified in the configuration file.

zyweven commented 2 years ago

Hi! Actually, the requirement for RAM is relatively high (around 30G), because we preload the whole dataset into the memory in our code. But this is avoidable at the cost of some additional training time (reading data from disk). Please change default value for preload argument in line 64 of FSC147_dataset.py from True to False. We will modify the code so that the preload choice for dataset can be modified in the configuration file.

thanks for your answer!:)