Clarify readme and fix filtering with transformers

README.md

@@ -21,7 +21,7 @@ Additionally:
 - Holding ctrl or cmd at any time will pause image processing temporarily, or you can specify keyboard combos in the config file to pause/unpause and switch the OCR provider from anywhere (refer to the config file or `owocr -h`)
 - You can enable notifications in the config file or with `-n` to show the text with a native OS notification. **Important for macOS users:** if you use Python from brew, you need to enter this command in your terminal before the first notification: `codesign -f -s - $(brew --cellar python)/3.*/Frameworks/Python.framework` (works on Ventura/Sonoma). Older macOS versions might require Python to be installed from the [official website](https://www.python.org/downloads/). Nothing can be done about this unfortunately.
 - Optionally, you can speed up the online providers by installing fpng-py: `pip install fpng-py` (requires setting up a developer environment on most operating systems/Python versions)
-- Optionally, you can improve filtering of non-Japanese text for screen capture by installing transformers: `pip install transformers`
+- Optionally, you can improve filtering of non-Japanese text for screen capture by installing transformers and sentencepiece: `pip install transformers sentencepiece`
 - A config file (which will be automatically created in `user directory/.config/owocr_config.ini`, on Windows `user directory` is the `C:\Users\yourusername` folder) can be used to configure the script, as an example to limit providers (to reduce clutter/memory usage) as well as specifying provider settings such as api keys etc. A sample config file is also provided [here](https://raw.githubusercontent.com/AuroraWright/owocr/master/owocr_config.ini)
 - For systems where text can be copied to the clipboard at the same time as images, if `*ocr_ignore*` is copied with an image, the image will be ignored (mostly useful for devs making their own sender tool)
 

owocr/run.py

@@ -294,8 +294,8 @@ class TextFiltering:
         if self.accurate_filtering:
             detection_results = self.pipe(new_blocks, top_k=2, truncation=True)
             for idx, block in enumerate(new_blocks):
-                if((detection_results[idx][0]['label'] == 'ja' and detection_results[idx][0]['score'] >= 0.85) or
-                   (detection_results[idx][1]['label'] == 'ja' and detection_results[idx][1]['score'] >= 0.85)):
+                if((detection_results[idx][0]['label'] == 'ja' and detection_results[idx][0]['score'] >= 0.20) or
+                   (detection_results[idx][1]['label'] == 'ja' and detection_results[idx][1]['score'] >= 0.20)):
                     final_blocks.append(block)
         else:
             for block in new_blocks: