-
First, I want to thank for this amazing library, I have been using it for a few months and I am completely amazed. I am using Unet and the results are pretty good. However, since I am having some issu…
-
My code is
```
loss = smp.utils.losses.BCELoss() \
+ smp.utils.losses.NoiseRobustDiceLoss(eps=1e-7) \
+ smp.utils.losses.TverskyLoss(eps=1e-7, alpha=0.7, beta=0.7)
```
Losses all…
-
I think `symmetric_focal_tversky_loss` needs to be partially changed from the code of the `asymmetric_focal_tversky_loss`
```python
# asymmetric_focal_tversky_loss
...
back_dice = (1-dice_clas…
-
Hi, I'm curious about the focal_tversky loss can be implemented into a multi class problem?
I saw the model is the sigmoid layer for binary class, how should I modify the loss?
-
Hi,
I am training an image segmentation network and using focal tversky loss, which comes with three parameters (alpha, gamma, smooth). Below is the code I am using to implement the loss:
```
tv…
-
Documentation doesn't represent the parameters and `asymmetric_focal_loss` is taking wrong attributes
```
def unified_focal_loss(weight=0.5, delta=0.6, gamma=0.2):
"""
:param weight: rep…
-
**- 版本、环境信息：**
1）PaddlePaddle版本：1.7
2）CPU：
3）GPU：Tesla V100
4）系统环境：AI Studio
- 训练信息
1）单机，单卡
2）显存信息 16G
**- 复现信息：**
损失函数：
def tversky(predict, label):
predict = …
-
Are there any plans to add an easy way to include "loss mask" to `Dataset` class for segmentation task? As what I understood, it only takes `img` and `seg`?
For example I have such CT volumes with …
-
Thank you for implementing a great tool. I recently tried using sweep inside a Kaggle kernel and noticed that it takes a lot longer to run 1 epoch than normal model training.
For example, when I…
-
I have an image segmentation task with small batch size (4-8) and some samples that have only the background (negative) class.
I have implemented lovasz softmax as below:
```python
loss2 = lova…