search

czczup / ViT-Adapter

[ICLR 2023 Spotlight] Vision Transformer Adapter for Dense Predictions
https://arxiv.org/abs/2205.08534
Apache License 2.0
1.23k stars 137 forks source link

img_norm_config for maskrcnn dinov2 large seems to be accidentally default #184

Open CardenB opened 3 weeks ago

CardenB commented 3 weeks ago

https://github.com/czczup/ViT-Adapter/blob/94ffa6b6134b99d838312e2e042d6ac3a52a7ef8/detection/configs/mask_rcnn/dinov2/mask_rcnn_dinov2_adapter_large_fpn_3x_coco.py#L47

has all 127.5 values rather than the values in small and base

czczup commented 3 weeks ago

Hello, this should be my typo, it may slightly damage the performance.

CardenB commented 3 weeks ago

Was it also intended to have image size 592 for small, base, and large variants?