mirrors/pytorch-image-models
1
0
Fork 0
mirror of https://github.com/huggingface/pytorch-image-models.git synced 2025-06-03 15:01:08 +08:00
Code Issues Projects Releases Wiki Activity

Fix positional embedding resampling for non-square inputs in ViT

Browse Source
This commit is contained in:
Wojtek Jasiński 2024-11-01 19:46:46 +01:00 committed by Ross Wightman
parent 51ac8d2efb
commit 3ae3f44288
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
timm/models/vision_transformer.py
View File

@ -669,9 +669,11 @@ class VisionTransformer(nn.Module):
if self.dynamic_img_size: if self.dynamic_img_size:
B, H, W, C = x.shape B, H, W, C = x.shape
prev_grid_size = self.patch_embed.grid_size
pos_embed = resample_abs_pos_embed( pos_embed = resample_abs_pos_embed(
self.pos_embed, self.pos_embed,
(H, W), new_size=(H, W),
old_size=prev_grid_size,
num_prefix_tokens=0 if self.no_embed_class else self.num_prefix_tokens, num_prefix_tokens=0 if self.no_embed_class else self.num_prefix_tokens,
) )
x = x.view(B, -1, C) x = x.view(B, -1, C)