SAM_WSSS
SAM_WSSS copied to clipboard
cskyl
Question about the requirement of GPU
Hi! How many GPUs and what kind of GPUs are required in your work? Thanks!
Hi, thank you for asking. You don't need any GPU to generate the pseudo labels, they can be fully produced using only CPUs. On a single process, the rate is around 100~200 imgs / minute on my laptop, by using multi process, it takes about 10mins to generate the pseudo label for PASCAL train_aug (~10k imgs).
Hi! How many GPUs and what kind of GPUs are required in your work? Thanks!
Hi, thank you for asking. You don't need any GPU to generate the pseudo labels, they can be fully produced using only CPUs. On a single process, the rate is around 100~200 imgs / minute on my laptop, by using multi process, it takes about 10mins to generate the pseudo label for PASCAL train_aug (~10k imgs).
Hi! How many GPUs and what kind of GPUs are required in your work? Thanks!
Hi! How many GPUs and how long are required for training semantic segmentation with pseudo labels? Thank you
Hi, thank you for asking. You don't need any GPU to generate the pseudo labels, they can be fully produced using only CPUs. On a single process, the rate is around 100~200 imgs / minute on my laptop, by using multi process, it takes about 10mins to generate the pseudo label for PASCAL train_aug (~10k imgs).
Hi! How many GPUs and what kind of GPUs are required in your work? Thanks!
Hi! How many GPUs and how long are required for training semantic segmentation with pseudo labels? Thank you
Hi! We are using 2 V100 GPUs to train the semantic segmentation models with our pseudo labels, more details about the training are following the repo of Deeplab Model https://github.com/kazuto1011/deeplab-pytorch. You can find the hyperparameters we used in our paper https://arxiv.org/pdf/2305.05803.pdf. Thank you for your comments!