feat: add yolo model for nutriscore object detection (#1356)

* chore: change OpenCV dependency

We're not using imshow function, so we can use the headless version of
OpenCV, which is lighter.
See https://pypi.org/project/opencv-python/ for more information.

* feat: implement YOLO models

* fix: fix UPC detection after OpenCV upgrade

* fix: disable UPC detection

Makefile

@@ -118,6 +118,8 @@ dl-object-detection-models:
 			mkdir -p $${dir}/1; \
 			wget -cO - https://github.com/openfoodfacts/robotoff-models/releases/download/$${asset_name}-1.0/model.onnx > $${dir}/1/model.onnx; \
 	done; \
+	mkdir -p nutriscore-yolo/1; \
+	wget -cO - https://huggingface.co/openfoodfacts/nutriscore-yolo/resolve/main/weights/best.onnx > nutriscore-yolo/1/model.onnx
 
 dl-category-classifier-model:
 	@echo "⏬ Downloading category classifier model files …"

models/triton/nutriscore-yolo/labels.txt

@@ -0,0 +1,5 @@
+nutriscore-a
+nutriscore-b
+nutriscore-c
+nutriscore-d
+nutriscore-e

pyproject.toml

@@ -57,7 +57,7 @@ influxdb-client = "~1.34.0"
 jsonschema = "~4.4.0"
 orjson = ">=3.8.2,<3.10.0"
 Pillow = ">=9.3,<10.4"
-numpy = "~1.23.5"
+numpy = "~1.26.4"
 protobuf = "~3.19.0"
 Pint = "0.22"
 APScheduler = "~3.10.1"
@@ -72,7 +72,7 @@ python-redis-lock = "~4.0.0"
 transformers = ">=4.30.2,<4.39.0"
 lark = "~1.1.4"
 h5py = "~3.8.0"
-opencv-contrib-python = "~4.7.0.72"
+opencv-contrib-python-headless = "~4.10.0.84"
 toml = "~0.10.2"
 openfoodfacts = "0.1.11"
 imagehash = "~4.3.1"

robotoff/insights/extraction.py

@@ -86,9 +86,9 @@ def run_object_detection_model(
 
     timestamp = datetime.datetime.utcnow()
     results = ObjectDetectionModelRegistry.get(model_name.value).detect_from_image(
-        image, output_image=False, triton_uri=triton_uri
+        image, output_image=False, triton_uri=triton_uri, threshold=threshold
     )
-    data = results.to_json(threshold=threshold)
+    data = results.to_json()
     max_confidence = max((item["score"] for item in data), default=None)
     return ImagePrediction.create(
         image=image_model,