Datasets:

ArXiv:
Tags:
art
License:
schirrmacher commited on
Commit
e7b05d2
·
verified ·
1 Parent(s): ea07c6e

Upload folder using huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +11 -9
README.md CHANGED
@@ -7,6 +7,8 @@ pretty_name: Human Segmentation Dataset
7
 
8
  # Human Segmentation Dataset
9
 
 
 
10
  This dataset was created **for developing the best fully open-source background remover** of images with humans.
11
  The dataset was crafted with [LayerDiffuse](https://github.com/layerdiffusion/LayerDiffuse), a Stable Diffusion extension for generating transparent images.
12
 
@@ -15,19 +17,10 @@ Since Stable Diffusion is not perfect, the dataset contains images with flaws. S
15
 
16
  The resulting model will be similar to [RMBG-1.4](https://huggingface.co/briaai/RMBG-1.4), but with open training data/process and commercially free to use.
17
 
18
- I had some trouble with the Hugging Face file upload. You can find the data here: [Google Drive](https://drive.google.com/drive/folders/1K1lK6nSoaQ7PLta-bcfol3XSGZA1b9nt?usp=drive_link).
19
-
20
  The dataset contains transparent images of humans (`/humans`) which are randomly combined with backgrounds (`/backgrounds`). Then the ground truth (`/gt`) for segmentation was computed based on the transparent images. The results are written to a training and validation dataset.
21
 
22
  I created more than 5.000 images with people and more than 5.000 diverse backgrounds.
23
 
24
- # Examples
25
-
26
- Here you can see an augmented image and the resulting ground truth:
27
-
28
- ![](example_image.png)
29
- ![](example_ground_truth.png)
30
-
31
  # Create Training Dataset
32
 
33
  The following script creates augmented training and validation data.
@@ -38,10 +31,19 @@ Notice: download the dataset from [Google Drive](https://drive.google.com/drive/
38
  ./create_dataset.sh
39
  ```
40
 
 
 
 
 
 
 
 
41
  # Support
42
 
43
  If you identify weaknesses in the data, please contact me.
44
 
 
 
45
  # Changelog
46
 
47
  - Added more diverse backgrounds (natural landscapes, streets, houses)
 
7
 
8
  # Human Segmentation Dataset
9
 
10
+ [>>> Download Here <<<](https://drive.google.com/drive/folders/1K1lK6nSoaQ7PLta-bcfol3XSGZA1b9nt?usp=drive_link).
11
+
12
  This dataset was created **for developing the best fully open-source background remover** of images with humans.
13
  The dataset was crafted with [LayerDiffuse](https://github.com/layerdiffusion/LayerDiffuse), a Stable Diffusion extension for generating transparent images.
14
 
 
17
 
18
  The resulting model will be similar to [RMBG-1.4](https://huggingface.co/briaai/RMBG-1.4), but with open training data/process and commercially free to use.
19
 
 
 
20
  The dataset contains transparent images of humans (`/humans`) which are randomly combined with backgrounds (`/backgrounds`). Then the ground truth (`/gt`) for segmentation was computed based on the transparent images. The results are written to a training and validation dataset.
21
 
22
  I created more than 5.000 images with people and more than 5.000 diverse backgrounds.
23
 
 
 
 
 
 
 
 
24
  # Create Training Dataset
25
 
26
  The following script creates augmented training and validation data.
 
31
  ./create_dataset.sh
32
  ```
33
 
34
+ # Examples
35
+
36
+ Here you can see an augmented image and the resulting ground truth:
37
+
38
+ ![](example_image.png)
39
+ ![](example_ground_truth.png)
40
+
41
  # Support
42
 
43
  If you identify weaknesses in the data, please contact me.
44
 
45
+ I had some trouble with the Hugging Face file upload. You can find the data here: [Google Drive](https://drive.google.com/drive/folders/1K1lK6nSoaQ7PLta-bcfol3XSGZA1b9nt?usp=drive_link).
46
+
47
  # Changelog
48
 
49
  - Added more diverse backgrounds (natural landscapes, streets, houses)