Small changes in swin_transformer.py for Image size scale

[zichengsaber]

Thanks for your remarkable work, when I use the pretrained model ，I found it difficult to scale the input Image size，even I use the image size like window_size*2^n , it loads failed. So I check the code, find that In Class SwinTransformerBlock code
self.register_buffer("attn_mask", attn_mask) is unnecessary，because attn_mask is not a learnable parameter, if you register it as buffer, it will hinder you change the input Image size

[zeliu98]

Hi, you can just delete the "attn_mask" in the pre-trained model. Besides, we have also provided a function for your purpose:

Swin-Transformer/utils.py

Line 46 in 5d2aede

def load_pretrained(config, model, logger):

[zichengsaber]

Hi, you can just delete the "attn_mask" in the pre-trained model. Besides, we have also provided a function for your purpose:

Swin-Transformer/utils.py

Line 46 in 5d2aede

def load_pretrained(config, model, logger):

thank you very much I will try it

[khanhnguyenuet]

Hi, you can just delete the "attn_mask" in the pre-trained model. Besides, we have also provided a function for your purpose:

Swin-Transformer/utils.py

Line 46 in 5d2aede

def load_pretrained(config, model, logger):

Hi there, I did as u said to load pretrained model but I only can load the state_dict to model and still can not parse the image through the model due to the lack of attn_mask. Could u explain why?. In case I want to train with another image size, how can I train the model?