You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README.md
+2-2
Original file line number
Diff line number
Diff line change
@@ -1,8 +1,8 @@
1
1
# PicsArtHack-binary-segmentation
2
2
3
3
The goal of the hackathon was to build some image processing algorithm which can be helpful for [PicsArt](https://picsart.com/?hl=en) applications.
4
-
Here I publish results of the first stage: segmenting people on selfies.
5
-
PicsArt gave labeled [dataset](https://drive.google.com/file/d/1_e2DcZnjufx35uSmQElN5mpdo-Rlv7ZI/view?usp=sharing). [Dice](https://en.wikipedia.org/wiki/S%C3%B8rensen%E2%80%93Dice_coefficient) coef. was used as evaluation metric.
4
+
Here I publish results of the first stage: segmenting people on selfies.
5
+
PicsArt gave us labeled dataset. [Dice](https://en.wikipedia.org/wiki/S%C3%B8rensen%E2%80%93Dice_coefficient) coef. was used as evaluation metric.
6
6
I noticed that a lot of images has been labeled by another segmentation model due to a lot of artifacts around the masks borders. Also in test dataset appears copies of train set images. So after training, I did not expect good results on images "from the wild".
0 commit comments