add improvements

vision_agent_tools/models/florence2.py

@@ -15,7 +15,7 @@
     Device,
     Florence2ResponseType,
     BaseMLModel,
-    ODResponse,
+    ODWithScoreResponse,
     Florence2OCRResponse,
     Florence2TextResponse,
     Florence2OpenVocabularyResponse,
@@ -360,7 +360,11 @@ def _serialize(
                 | PromptTask.REGION_PROPOSAL
             ):
                 detections.append(
-                    ODResponse(bboxes=detection["bboxes"], labels=detection["labels"])
+                    ODWithScoreResponse(
+                        bboxes=detection["bboxes"],
+                        labels=detection["labels"],
+                        scores=[1.0] * len(detection["labels"]),
+                    )
                 )
             case PromptTask.OCR_WITH_REGION:
                 detections.append(

vision_agent_tools/models/sam2.py

@@ -79,8 +79,8 @@ class Sam2(BaseMLModel):
     """It receives images, a prompt and returns the instance segmentation for the
     text prompt in each frame."""
 
-    def __init__(self, model_config: Sam2Config | None = Sam2Config()):
-        self.model_config = model_config
+    def __init__(self, model_config: Sam2Config | None = None):
+        self.model_config = model_config or Sam2Config()
         self.image_model = SAM2ImagePredictor.from_pretrained(
             self.model_config.hf_model
         )

vision_agent_tools/tools/text_to_object_detection.py

@@ -33,6 +33,12 @@ class TextToObjectDetectionRequest(BaseModel):
         le=1.0,
         description="The IoU threshold value used to apply a dummy agnostic Non-Maximum Suppression (NMS).",
     )
+    chunk_length_frames: int | None = Field(
+        default=None,
+        ge=1,
+        le=30,
+        description="The number of frames for each chunk of video to analyze. The last chunk may have fewer frames.",
+    )
     confidence: float | None = Field(
         default=None,
         ge=0.0,
@@ -62,7 +68,7 @@ class TextToObjectDetection(BaseTool):
     def __init__(
         self,
         model: TextToObjectDetectionModel = TextToObjectDetectionModel.OWLV2,
-        model_config: OWLV2Config | None = None,
+        model_config: OWLV2Config | Florence2Config | None = None,
     ):
         self.model_name = model
 
@@ -83,6 +89,7 @@ def __call__(
         video: VideoNumpy | None = None,
         *,
         nms_threshold: float = 0.3,
+        chunk_length_frames: int | None = None,
         confidence: float | None = None,
     ) -> list[dict[str, Any]]:
         """Run object detection on the image based on text prompts.
@@ -96,6 +103,9 @@ def __call__(
                 A numpy array containing the different images, representing the video.
             nms_threshold:
                 The IoU threshold value used to apply a dummy agnostic Non-Maximum Suppression (NMS).
+            chunk_length_frames:
+                The number of frames for each chunk of video to analyze.
+                The last chunk may have fewer frames.
             confidence:
                 Confidence threshold for model predictions.
 
@@ -108,6 +118,7 @@ def __call__(
             images=images,
             video=video,
             nms_threshold=nms_threshold,
+            chunk_length_frames=chunk_length_frames,
             confidence=confidence,
         )
 
@@ -137,6 +148,7 @@ def __call__(
                 images=images,
                 video=video,
                 nms_threshold=nms_threshold,
+                chunk_length_frames=chunk_length_frames,
             )
 
     def to(self, device: Device):

vision_agent_tools/tools/text_to_object_detection_sam2.py

@@ -23,8 +23,8 @@
 
 
 class Text2ODSAM2Config(BaseModel):
-    sam2_config: Sam2Config | None = Sam2Config()
-    text2od_config: Florence2Config | OWLV2Config | None = OWLV2Config()
+    sam2_config: Sam2Config | None = None
+    text2od_config: Florence2Config | OWLV2Config | None = None
 
 
 class Text2ODSam2Request(BaseModel):
@@ -88,14 +88,14 @@ class Text2ODSAM2(BaseMLModel):
     def __init__(
         self,
         model: TextToObjectDetectionModel = TextToObjectDetectionModel.OWLV2,
-        model_config: Text2ODSAM2Config | None = Text2ODSAM2Config(),
+        model_config: Text2ODSAM2Config | None = None,
     ):
         """
         Initializes the Text2ODSAM2 object with a pre-trained text2od model
         and a SAM2 model.
         """
         self._model = model
-        self._model_config = model_config
+        self._model_config = model_config or Text2ODSAM2Config()
         self._text2od = TextToObjectDetection(
             model=model, model_config=self._model_config.text2od_config
         )
@@ -170,7 +170,9 @@ def __call__(
 
         text2od_payload_response = self._text2od(**text2od_payload)
         od_response = [
-            ODWithScoreResponse(**item) if len(item.get("labels")) > 0 else None
+            ODWithScoreResponse(**item)
+            if item is not None and len(item.get("labels")) > 0
+            else None
             for item in text2od_payload_response
         ]
         if images is not None: