AFeng

Hi, we conducted experiments using 8 V100 GPUs. The total batch size was set to 1024, and the training took approximately 2 and a half days. When we trained for...

Hi, here are two simple ways u can try: (1) reduce the number of channel (eg. 256->128) (2) reduce the number of block (eg. 12->6) Also, you need to confirm...

Hi! Apologies for the delayed response. We plan to release a portion of the dataset (including controllable generation) by May 2025—please stay tuned for updates.

Thank you for your interest! We initially attempted to use vanilla convolution, but we found that it led to a larger number of parameters in the module. As a result,...

Hi, maybe in SMT/config.py the line 3 should be "from yacs.config import CfgNode as CN" instead of "from config import CfgNode as CN"

hi! You can delete these unrelated codes, for example: from .build import build_loader as _build_loader def build_loader(config, simmim=False, is_pretrain=False): if not simmim: return _build_loader(config)

你好，可以的，我们在后续完善的时候会放出coco的模型权重，请随时关注

Hello, our SMT-T model should have a higher throughput speed than CrossFormer-S (833 vs. 672 images/s). Theoretically, the training time should not differ significantly. One possible reason is that depth-wise...

The impact on training speed is indeed a drawback of this model. Thank you for bringing it to our attention. If longer training times are not feasible for you, we...

Hi, Thank you for your comment and for bringing up the excellent MaxViT work. I am impressed by the innovative ideas and the results presented in the paper. In the...