fix

Author: jonafeucht

README.md

@@ -6,10 +6,15 @@ Fine-tune a pre-trained **TF-EfficientNet-Lite4** model on a custom image datase
 2. Create subfolders (example: dog, cat, fish, bird) and populate them.
 3. Run `python src/fine_tune_from_base.py`.
 
-When its finished it will return a checkpoint file, labels.txt, and the fine-tuned model.
+When its finished it will return a checkpoint file.
+
+### Create a tflite model
+1. Run `python src/convert_to_tflite.py`.
+
+This will return a tflite and labels.txt file.
 
 > [!NOTE]
-> Fine-tuned models are using in conjunction with the [Vision Classification](https://github.com/doppeltilde/vision_classification) project, utilizing its Mediapipe Image Classification task.
+> Fine-tuned models are used in conjunction with the [Vision Classification](https://github.com/doppeltilde/vision_classification) project, utilizing its Mediapipe Image Classification task.
 
 ---
 _Notice:_ _This project was initally created to be used in-house, as such the

requirements.txt

@@ -7,4 +7,5 @@ tqdm==4.67.3
 pillow==12.2.0
 numpy==2.4.4
 mediapipe==0.10.33
-flatbuffers==25.12.19
+flatbuffers==25.12.19
+tflite-support==0.4.4

src/convert_to_tflite.py

@@ -7,6 +7,8 @@
 import traceback
 from pathlib import Path
 import flatbuffers
+from tflite_support import metadata as _metadata
+from tflite_support import metadata_schema_py_generated as _metadata_fb
 
 script_dir = Path(__file__).parent.resolve()