sallymmx/transvos

This is an official implementation of TransVOS

transvos's Issues

Results without pre-training

Hi, thanks for releasing the code!

Could you provide the results without pre-training?

Which version of coco dataset do you use?

2014?

The config about the pretraing is the same as the main training?

release test code?

sorry i can't find test code,can you please release your test code? thank you

Out of memory

Hi, we follow your default config and train on V100 with 32GB. However, it raise out of memory error. We have to reduce the batch size from 4 to 1 in order to fit in the gpu memory. I wonder if there is any mismatch between the code in the repo and the one you used to train on TITAN RTX which has 24GB gpu memory? Thanks

config.py for pretrain

Hi, can you please share your config.py for pretrain? We used the same config from the repo for pretrain but got 80.15 accuracy. Much lower than your released code.

How many days does the pretraining take?

Is 1.5 days the total time of the two stages of training?

The evaluation curve on training is not gradually rize to the values in the paper.

My GPUs are poor, and I set the batchsize to 2, num_workers to 4. The other configs are same to default.
When training the J&F rize to top with 65 epochs, and then it towards to lower. Is it normal？
Training only with DAVIS17, the top J&F only 72. Is it caused by the half batchsize?
Thank you!!

release the pre-train model?

hi, Pre-training process may take days to weeks, can you please release the pre-train model on static images?