Running results

[Biyb]

I have two questions for the author:

First, where are the results such as mIOU and mDice calculated? In Issue #16, you also provided the code. However, due to my limited coding ability, I'm not sure which part of the existing codebase the code you provided should be inserted into.

Second, in the text.py file, what paths should be stored in the two parameters --checkpoint and --test_gt_path?

I would greatly appreciate it if you could help me out. Thank you!

[xiongxyowo]

Hi, in general, the calculation of Dice and IoU can be divided into the following situations:

1. The IoU of foreground lesions/organs and the ground truth are calculated after binarizing the prediction results with a confidence of 0.5. The result is usually denoted as "IoU", which corresponds to the code in calculate evaluation metrics such as Dice, IOU, and HD #16 .
2. Binarize the prediction results with different confidence levels, and then calculate the average under different confidence levels. The result is usually denoted as "mIoU", which corresponds to the code in eval.py.
3. For multi-class segmentation tasks, calculate the average IoU of different classes. In this case, the result is also denoted as "mIoU", but our code does not yet support this calculation, you can modify it yourself.

---

同学你好，一般来说，dice与iou的计算分为如下几种情况：

1. 将预测结果使用0.5的置信度二值化后计算前景病灶/器官与真值的iou。此时结果一般记为IoU，对应着#16中的代码
2. 将预测结果使用不同的置信度进行二值化，然后计算不同置信度下的平均。此时结果一般记为mIoU，对应eval.py中的代码
3. 对于多类分割任务，计算不同前景类别iou的平均。此时结果也一般记为mIoU，不过我们的SAM2-UNet代码尚未支持这种计算方式，你可以自行进行修改。

[Biyb]

这是来自QQ邮箱的假期自动回复邮件。   你好，我最近正在休假中，无法亲自回复你的邮件。我将在假期结束后，尽快给你回复。

[xiongxyowo]

checkpoint: pth file obtained after training，such as /your_dir/SAM2-UNet-20.pth
test_gt_path: the directory of the ground truth of testset，such as /your_dir/masks/

---

checkpoint: 训练完毕后得到的pth文件，例如/your_dir/SAM2-UNet-20.pth
test_gt_path: 测试集真值的文件夹目录，例如/your_dir/masks/