Zhaochong An

> Could you check whether `torch.nn.SyncBatchNorm` has this method? I just check as follows: $ torch.nn.SyncBatchNorm $ torch.nn.SyncBatchNorm._specify_ddp_gpu_num() Traceback (most recent call last): File "", line 1, in TypeError: _specify_ddp_gpu_num()...

But my current codebase is from mmseg which has assert clause to check the mmcv version to be from 1.1.4 to 1.3.0 in its __init__.py. ``` MMCV_MIN = '1.1.4' MMCV_MAX...

Moreover, can you release the pre-process code? So people can train the model on other dataset. Thank you so much!

Hi, I am using 4 gpus for the training. The issue seems related to your wrong installation of torch https://discuss.pytorch.org/t/runtimeerror-distributed-package-doesnt-have-nccl-built-in/176744.

Hi, thank you! Could you please try setting the training amp type as bfloat16 at: https://github.com/ZhaochongAn/COSeg/blob/6116345a32b27f4d97c4aa61a7a21f58f8ff885e/main_fs.py#L607 by using: `with torch.cuda.amp.autocast(enabled=use_amp, dtype=torch.bfloat16)`. In our mulimodality codebase, we use the bfloat16 type...

Hi, in our experiments, we only compare to the previous methods that release the code. Based on their code, we retrain the model in our setting and use our visualization...

For other networks you want to evaluate, you could write their model file by following the input and output format in the forward function of coseg.py (https://github.com/ZhaochongAn/COSeg/blob/main/model/coseg.py). Then, you could...

Hey, for the dgcnn exps, we are conducting the pretraining and episode training using this codebase by adding a dgcnn backbone file. Other parts are kept same. So, you just...

Feel free to reopen it for new questions. Close it for now.