search

Hzzone / ProPos

Self-Supervised Learning for Deep Clustering (TPAMI 2022)
104 stars 15 forks source link

Training speed of ImageNet-10 #9

Closed kasteric closed 1 year ago

kasteric commented 1 year ago

Thanks for your great code bases and fast responses! i just have a minor question regarding the training speed of imagenet-10/imagenet-dogs, which i find is very slow in dataloader processing, causing low gpu utilization. Sorry I am not very familiar with training ImageNet, is this a issue that can be improved or a common phenonmenon? I really appreciate your help!

Hzzone commented 1 year ago

Try to use an SSD disk to store your images.

kasteric commented 1 year ago

Thanks for your generous response, I found in the paper the training time of CIFAR10 is around 10 hours while that of ImageNet-10 is only 1 hour. However, in my experiments, it took just 3 hours to train CIFAR10 (byol) on 4 RTX 3090 cards for 1000 epochs but it took me more than 5 hours (maybe even longer) to train ImageNet-10, which is very strange, are you using ImageFolder for ImageNet-10?