Balachandar Gowrisankar

Multi-GPU training command starts same job in multiple GPUs instead of distributed training

1

Hi, I followed the mmseg tutorial to set up multi-GPU(2 GPUs) training for a job. But instead of using 2 GPUs to train, it starts the same job individually in...

Keeping other parameters unchanged, if I try to use SGD instead of AdamW optimizer for Segformer model(modified with a Swin backbone), the results are so bad. I do get decent...

Keeping other parameters unchanged, if I try to use SGD instead of AdamW optimizer for Segformer model(modified with a Swin backbone), the results are so bad. I do get decent...

Resolution mismatch between tampered images and masks

1

Hi, Thank you for open-sourcing the dataset. The resolutions of the forged images and masks are not matching for about 9 cases. May I know how that can be resolved?

Default weight initialization

1

Hi, If I don't specify any weight initialization method in init_cfg or implement a custom init_weights() function, how does mmseg assign the initial weights? What is the default method used...