Phips commited on
Commit
f4ae03a
·
verified ·
1 Parent(s): 6874ebb

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +19 -5
README.md CHANGED
@@ -3,10 +3,20 @@ license: cc-by-4.0
3
  ---
4
  # BHI SISR Dataset
5
 
6
- The BHI SISR Dataset's purpose is for training single image super-resolution models and is a result of tests on my BHI filtering method, which I made [a huggingface community blogpost about](https://huggingface.co/blog/Phips/bhi-filtering).
7
 
8
- TODO it consists of X images, which are all 512x512px dimensions and in the png format.
9
- TODO visual example of the dataset
 
 
 
 
 
 
 
 
 
 
10
 
11
  ## Used Datasets
12
 
@@ -66,7 +76,7 @@ HFA2K -> 2'280 Tiles
66
  ModernAnimation1080_v3 -> 4'109 Tiles
67
  Nomos_Uni -> 2'466 Tiles
68
  Nomosv2 -> 5'226 Tiles
69
- inaturalist_2019 -> 131'943 Tiles
70
 
71
 
72
  ## Files
@@ -77,4 +87,8 @@ Files have been named with '{dataset_name}_{index}.png' so that if one of these
77
 
78
  ## Optimization
79
 
80
- Then I used [oxipng](https://github.com/shssoichiro/oxipng) ("oxipng --strip safe --alpha *.png") for optimization.
 
 
 
 
 
3
  ---
4
  # BHI SISR Dataset
5
 
6
+ The BHI SISR Dataset's purpose is for training single image super-resolution models and is a result of tests on my BHI filtering method, which I made [a huggingface community blogpost about](https://huggingface.co/blog/Phips/bhi-filtering), which can be extremely summarized by that removing (by filtering) only the worst quality tiles from a training set has a way bigger positive effect on training metrics than keeping only the best quality training tiles.
7
 
8
+ It consists of 390'241 images, which are all 512x512px dimensions and in the webp format.
9
+
10
+ The advantage of such a big dataset is when applying degradations in a randomized manner to create a corresponding LR, the distribution of degradations and strenghts should be sufficient because of the quantity of training tiles. I will create some corresponding x4 LR datasets to this one and publish them aswell.
11
+ Though if an on-the-fly degradation pipeline is used during training, such a high quantity of training tiles would probably generally not be needed since longer training iterations make sure of distribution.
12
+
13
+ Size on disc:
14
+ ```
15
+ du BHI_HR
16
+ 131199816 BHI_HR/
17
+ ```
18
+
19
+ Also for the future, I am releasing the full dataset here. But there can of course be attempts in the future to make distilled versions of this dataset that perform better since I might find additional metrics or filtering methods in the future that might help reduce dataset size while achieving better training validation metric performance.
20
 
21
  ## Used Datasets
22
 
 
76
  ModernAnimation1080_v3 -> 4'109 Tiles
77
  Nomos_Uni -> 2'466 Tiles
78
  Nomosv2 -> 5'226 Tiles
79
+ inaturalist_2019 -> 131'943 Tiles
80
 
81
 
82
  ## Files
 
87
 
88
  ## Optimization
89
 
90
+ Then I used [oxipng](https://github.com/shssoichiro/oxipng) ("oxipng --strip safe --alpha *.png") for optimization.
91
+
92
+ ## WebP conversion
93
+
94
+ The files have then been converted to lossless webp simply to save storage space locally and for faster uploading/downloading here on huggingface. This reduced size of the dataset by around 50 GB.