Shadow detection结果不如预期

[NnAracy]

作者你好，感谢分享你们的工作。我按照README以如下指令执行demo后得到了异常的结果

python demo/demo.py--config-file cconfigs/sd/sd_dinov2_large.yaml\
  --input focus/data/test/camera-rgb_87557869742675.jpg \
  --output output \
  --opts MODEL.WEIGHTS weights/focus_large_sd.pth

原图

Result

这个dataset的图片来自鱼眼镜头，不确定是不是外圈黑边影响到了model，想听听看你的意见，谢谢。

[geshang777]

Hi, this issue is likely caused by the distribution gap between your test image and the ISTD dataset we used to train our shadow detection model, see #2 . If you have a fisheye-lens shadow detection dataset, you can fine-tune on top of our pretrained weights.