Update MZKBlank description

Author: v-dvorak

README.md

@@ -11,6 +11,7 @@ Semester work at MFF CUNI under the guidance of Jirka Mayer.
 - [User documentation](docs/README.md)
 - [Download the latest dataset](https://github.com/v-dvorak/omr-layout-analysis/releases/tag/Latest) (~2.5 GB)
 - [Download the latest model](https://github.com/v-dvorak/omr-layout-analysis/releases/tag/Models) (~50 MB)
+- [MZKBlank dataset](app/MZKBlank/README.md)
 
 ## Dataset overview
 
@@ -109,6 +110,7 @@ System are mainly created by looking at system staves at approximately the same
 - **September 2024**
   - added random shuffling to train/test split
   - added `seed` option
+  - unpublished MZKBlank
 
 - **August 2024**
   - complete dataset made public in Releases

app/MZKBlank/README.md

@@ -1,35 +1,24 @@
-# Negative samples (Background images)
+# MZKBlank (Background images)
 
 ## From the YOLO docs
 
 > **Background images**. Background images are images with no objects that are added to a dataset to reduce False
 > Positives (FP). We recommend about 0-10% background images to help reduce FPs (COCO has 1000 background images for
 > reference, 1% of the total). No labels are required for background images.
 
-The final dataset will consist of approximately ten thousand images, that means that we'll need around one thousand
-background images.
+We plan for the dataset to consist of approximately ten thousand images, that means that we'll need around one thousand background images.
 
 ## Where do we get the images from?
 
-From Moravská zemská knihovna - MZK.
-
-Using
-the [MZK search engine](https://www.digitalniknihovna.cz/mzk/search?access=open&licences=public&doctypes=sheetmusic) we
-can search specifically for open and public sheet music, there are thousands of them. MZK implements the really nice and
-convenient to use [IIIF API](https://iiif.io/api/image/3.0/). There are only two problems with it:
-
-- I was not able to find any documentation on how to use this api to search for documents using filters. (Like "give me
-  IDs to all sheet music documents".)
-
-- Page labelling is not consistent and some page IDs end up raising 4xx or 5xx errors.
-
-### [How does MZKScraper work?](https://github.com/v-dvorak/mzkscraper/blob/main/docs/README.md)
+From Moravská zemská knihovna (MZK). Using
+the [MZK search engine](https://www.digitalniknihovna.cz/mzk/search?access=open&licences=public&doctypes=sheetmusic) we can search specifically for open and public sheet music, there are thousands of them. MZK implements the really nice and convenient to use [IIIF API](https://iiif.io/api/image/3.0/). But it does implement all features that I would like to use, so I created [MZKScraper](https://github.com/v-dvorak/mzkscraper/blob/main/docs/README.md).
 
 ## Numbers of representative images
 
-The number of representative images for each label was counted as $\left\lceil\frac{\#\text{ samples of label} \:
-\cdot \: \#\text{ images wanted}}{\#\text{ samples in total}}\right\rceil$, this ensures that the number of
-representative images in total is around the value we want (1000 images).
+The number of representative images for each label was counted as 
+`round_up(total samples of a label * images wanted / samples in total)`, so that the ratios between images inside the dataset are approximately same as in all the library.
+
+All used images are cited [here](./docs/citations.txt).
 
 For quick image overview see [image grids](./docs/README.md).
 
@@ -49,6 +38,20 @@ For quick image overview see [image grids](./docs/README.md).
 | title page               |   212    |
 | **Total**                | **1006** |
 
+## Downloading the dataset
+
+Due to possible copyright issues, only page ids are publicly available and the user has to download them themselves.
+
+```bash
+# download
+python3 -m app.MZKBlank.download
+
+# create annotations (all will be emmpty)
+python3 -m app.MZKBlank.build
+```
+
+> Make sure, that you have reliable internet connection. Downloading images from the MZK might take some time.
+
 ## Labels provided by MZK
 
 ### Accepted labels