Add README.md

README.md

@@ -0,0 +1,24 @@
+# Image AutoML with AutoGluon, Sign Identification
+
+Dataset: ecopus/sign_identification
+
+Task: multiclass image classification using AutoGluon Multimodal.
+
+## Training
+- Preset: medium_quality
+- Backbone: timm_image, resnet18
+- Split: augmented 80 percent train, 20 percent test
+- External validation: original split
+
+## Results
+- Augmented test accuracy: 1.0000
+- Augmented test weighted F1: 1.0000
+- Original external accuracy: 1.0000
+- Original external weighted F1: 1.0000
+
+## How to load
+```python
+from autogluon.multimodal import MultiModalPredictor
+predictor = MultiModalPredictor.load('predictor_dir')
+pred = predictor.predict({'image': ['/path/to/img.png']})
+```