AFeng

Hi, sorry for not responding to you promptly. We did not process the captions; and we used the original benchmark without any modification.

Hi! Sorry for the delayed reply. The value ​**ratio=1.502145922746781** is calculated from w and h, representing the ​aspect ratio (width-to-height ratio). In the YAML file, however, ​ratio refers to ​data...

Hi, since a portion of the dataset is composed of in-house data, we will not be open-sourcing the entire dataset for now. However, we will provide a detailed data preparation...

hi, you can check the example in (https://github.com/AFeng-x/SMT/tree/main/visualiza/vis.ipynb).

Hello, the upernet_ressa_s.py should be the same as upernet_smt_b.py and upernet_smt_s.py in the same folder. You can first follow the guidance in the mmseg repo documentation to run the official...

Yes, it's feasible from a design perspective. However, due to the limited availability of GPU machines at the moment, we may temporarily be unable to complete the training of larger...

Hi, the difference between srun and torchrun is that srun relies on the SLURM system to allocate resources and launch parallel tasks in a cluster, typically used for institutionally deployed...

Hi, please keep an eye on our repo within the next week—the dataset will be released no later than May 6th.

Hi, our model requires a minimum of 8xA6000 (48GB) GPUs for training. The more and better GPUs, the faster the overall training speed will be. For inference and testing, only...

你好，请到https://github.com/AFeng-x/SMT?tab=readme-ov-file#main-results-on-imagenet-with-pretrained-models这的表格里面下载。（其中github就是下载路径）