kkakkkka / etris Goto Github PK

Hello.

Thank you for your nice work and code.

Do you have any code available for the visualization?

It will be helpful for my research.

Thank you!

Is the default number of bridger three?

Hello,

can you release pre-trained weights of ETRIS ?

Thanks

@kkakkkka

Nice work !

As reported in paper, oIoU is the main metric. But, the released code computes the mIoU, which is similar to CRIS (also see issue). So, this inconsistency bothering me.

Does the test set used in the experiment combine testA and testB? Is the IOU result output by test.py the mean value of testA and testB?

Hi, @kkakkkka , I get stuck with 2 GPU when training. But I can train it successfully with 1 GPU. Do you have any suggestions? Lots of thanks!

Details：I only modify the gpu in the train.sh and gte stuck here.

File "train.py", line 62, in main mp.spawn(main_worker, nprocs=args.ngpus_per_node, args=(args, ))

Hi, thanks for your excellent work!
I run your code but got the error following:
The shape of the 2D attn_mask is torch.Size([17, 17]), but should be (77, 77).
How can I solve it ?

Excuse me, I want to segment an image by the text I input. Do I need to redesign an algorithm? Or is this code already in the project？

Hi,

Thanks for your great work.
I ran the code with ResNet-50 on RefCOCO and got the results 69.75/69.07/69.46 on val/test A/test B. The result on test A is much worse than the reported result 73.11.

Could you give some suggestions on this situation?

Many thanks.

Hi @zhjohnchan @kkakkkka ,

Could you please let me know how to train this model for a different dataset?