vision-agent 0.2.226__tar.gz → 0.2.228__tar.gz

{vision_agent-0.2.226 → vision_agent-0.2.228}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: vision-agent
-Version: 0.2.226
+Version: 0.2.228
 Summary: Toolset for Vision Agent
 Author: Landing AI
 Author-email: dev@landing.ai

{vision_agent-0.2.226 → vision_agent-0.2.228}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "poetry.core.masonry.api"
 
 [tool.poetry]
 name = "vision-agent"
-version = "0.2.226"
+version = "0.2.228"
 description = "Toolset for Vision Agent"
 authors = ["Landing AI <dev@landing.ai>"]
 readme = "README.md"

{vision_agent-0.2.226 → vision_agent-0.2.228}/vision_agent/tools/__init__.py

@@ -63,6 +63,10 @@ from .tools import (
     video_temporal_localization,
     vit_image_classification,
     vit_nsfw_classification,
+    custom_object_detection,
+    agentic_object_detection,
+    agentic_sam2_instance_segmentation,
+    agentic_sam2_video_tracking,
 )
 
 __new_tools__ = [

{vision_agent-0.2.226 → vision_agent-0.2.228}/vision_agent/tools/tools.py

@@ -290,6 +290,21 @@ def od_sam2_video_tracking(
             )
             function_name = "florence2_object_detection"
 
+        elif od_model == ODModels.AGENTIC:
+            segment_results = agentic_object_detection(
+                prompt=prompt,
+                image=segment_frames[frame_number],
+                fine_tune_id=fine_tune_id,
+            )
+            function_name = "agentic_object_detection"
+
+        elif od_model == ODModels.CUSTOM:
+            segment_results = custom_object_detection(
+                deployment_id=fine_tune_id,
+                image=segment_frames[frame_number],
+            )
+            function_name = "custom_object_detection"
+
         else:
             raise NotImplementedError(
                 f"Object detection model '{od_model}' is not implemented."
@@ -1217,6 +1232,139 @@ def countgd_visual_prompt_object_detection(
     return bboxes_formatted
 
 
+def custom_object_detection(
+    deployment_id: str,
+    image: np.ndarray,
+    box_threshold: float = 0.1,
+) -> List[Dict[str, Any]]:
+    """'custom_object_detection' is a tool that can detect instances of an
+    object given a deployment_id of a previously finetuned object detection model.
+    It is particularly useful when trying to detect objects that are not well detected by generalist models.
+    It returns a list of bounding boxes with normalized
+    coordinates, label names and associated confidence scores.
+
+    Parameters:
+        deployment_id (str): The id of the finetuned model.
+        image (np.ndarray): The image that contains instances of the object.
+        box_threshold (float, optional): The threshold for detection. Defaults
+            to 0.1.
+
+    Returns:
+        List[Dict[str, Any]]: A list of dictionaries containing the score, label, and
+            bounding box of the detected objects with normalized coordinates between 0
+            and 1 (xmin, ymin, xmax, ymax). xmin and ymin are the coordinates of the
+            top-left and xmax and ymax are the coordinates of the bottom-right of the
+            bounding box.
+
+    Example
+    -------
+        >>> custom_object_detection("abcd1234-5678efg", image)
+        [
+            {'score': 0.49, 'label': 'flower', 'bbox': [0.1, 0.11, 0.35, 0.4]},
+            {'score': 0.68, 'label': 'flower', 'bbox': [0.2, 0.21, 0.45, 0.5]},
+            {'score': 0.78, 'label': 'flower', 'bbox': [0.3, 0.35, 0.48, 0.52]},
+            {'score': 0.98, 'label': 'flower', 'bbox': [0.44, 0.24, 0.49, 0.58]},
+        ]
+    """
+    image_size = image.shape[:2]
+    if image_size[0] < 1 or image_size[1] < 1:
+        return []
+
+    files = [("image", numpy_to_bytes(image))]
+    payload = {
+        "deployment_id": deployment_id,
+        "confidence": box_threshold,
+    }
+    detections: List[List[Dict[str, Any]]] = send_inference_request(
+        payload, "custom-object-detection", files=files, v2=True
+    )
+
+    bboxes = detections[0]
+    bboxes_formatted = [
+        {
+            "label": bbox["label"],
+            "bbox": normalize_bbox(bbox["bounding_box"], image_size),
+            "score": bbox["score"],
+        }
+        for bbox in bboxes
+    ]
+    display_data = [
+        {
+            "label": bbox["label"],
+            "bbox": bbox["bounding_box"],
+            "score": bbox["score"],
+        }
+        for bbox in bboxes
+    ]
+
+    _display_tool_trace(
+        custom_object_detection.__name__,
+        payload,
+        display_data,
+        files,
+    )
+    return bboxes_formatted
+
+
+def custom_od_sam2_video_tracking(
+    deployment_id: str,
+    frames: List[np.ndarray],
+    chunk_length: Optional[int] = 10,
+) -> List[List[Dict[str, Any]]]:
+    """'custom_od_sam2_video_tracking' is a tool that can segment multiple objects given a
+    custom model with predefined category names.
+    It returns a list of bounding boxes, label names,
+    mask file names and associated probability scores.
+
+    Parameters:
+        deployment_id (str): The id of the deployed custom model.
+        image (np.ndarray): The image to ground the prompt to.
+        chunk_length (Optional[int]): The number of frames to re-run florence2 to find
+            new objects.
+
+    Returns:
+        List[Dict[str, Any]]: A list of dictionaries containing the score, label,
+            bounding box, and mask of the detected objects with normalized coordinates
+            (xmin, ymin, xmax, ymax). xmin and ymin are the coordinates of the top-left
+            and xmax and ymax are the coordinates of the bottom-right of the bounding box.
+            The mask is binary 2D numpy array where 1 indicates the object and 0 indicates
+            the background.
+
+    Example
+    -------
+        >>> custom_od_sam2_video_tracking("abcd1234-5678efg", frames)
+        [
+            [
+                {
+                    'label': '0: dinosaur',
+                    'bbox': [0.1, 0.11, 0.35, 0.4],
+                    'mask': array([[0, 0, 0, ..., 0, 0, 0],
+                        [0, 0, 0, ..., 0, 0, 0],
+                        ...,
+                        [0, 0, 0, ..., 0, 0, 0],
+                        [0, 0, 0, ..., 0, 0, 0]], dtype=uint8),
+                },
+            ],
+            ...
+        ]
+    """
+
+    ret = od_sam2_video_tracking(
+        ODModels.CUSTOM,
+        prompt="",
+        frames=frames,
+        chunk_length=chunk_length,
+        fine_tune_id=deployment_id,
+    )
+    _display_tool_trace(
+        custom_od_sam2_video_tracking.__name__,
+        {},
+        ret["display_data"],
+        ret["files"],
+    )
+    return ret["return_data"]  # type: ignore
+
+
 def qwen2_vl_images_vqa(prompt: str, images: List[np.ndarray]) -> str:
     """'qwen2_vl_images_vqa' is a tool that can answer any questions about arbitrary
     images including regular images or images of documents or presentations. It can be
@@ -2000,6 +2148,242 @@ def siglip_classification(image: np.ndarray, labels: List[str]) -> Dict[str, Any
     return response
 
 
+# agentic od tools
+
+
+def _agentic_object_detection(
+    prompt: str,
+    image: np.ndarray,
+    image_size: Tuple[int, ...],
+    image_bytes: Optional[bytes] = None,
+    fine_tune_id: Optional[str] = None,
+) -> Dict[str, Any]:
+    if image_bytes is None:
+        image_bytes = numpy_to_bytes(image)
+
+    files = [("image", image_bytes)]
+    payload = {
+        "prompts": [s.strip() for s in prompt.split(",")],
+        "model": "agentic",
+    }
+    metadata = {"function_name": "agentic_object_detection"}
+
+    if fine_tune_id is not None:
+        landing_api = LandingPublicAPI()
+        status = landing_api.check_fine_tuning_job(UUID(fine_tune_id))
+        if status is not JobStatus.SUCCEEDED:
+            raise FineTuneModelIsNotReady(
+                f"Fine-tuned model {fine_tune_id} is not ready yet"
+            )
+
+        # we can only execute fine-tuned models with florence2
+        payload = {
+            "prompts": payload["prompts"],
+            "jobId": fine_tune_id,
+            "model": "florence2",
+        }
+
+    detections = send_task_inference_request(
+        payload,
+        "text-to-object-detection",
+        files=files,
+        metadata=metadata,
+    )
+
+    # get the first frame
+    bboxes = detections[0]
+    bboxes_formatted = [
+        {
+            "label": bbox["label"],
+            "bbox": normalize_bbox(bbox["bounding_box"], image_size),
+            "score": bbox["score"],
+        }
+        for bbox in bboxes
+    ]
+    display_data = [
+        {
+            "label": bbox["label"],
+            "bbox": bbox["bounding_box"],
+            "score": bbox["score"],
+        }
+        for bbox in bboxes
+    ]
+    return {
+        "files": files,
+        "return_data": bboxes_formatted,
+        "display_data": display_data,
+    }
+
+
+def agentic_object_detection(
+    prompt: str,
+    image: np.ndarray,
+    fine_tune_id: Optional[str] = None,
+) -> List[Dict[str, Any]]:
+    """'agentic_object_detection' is a tool that can detect and count multiple objects
+    given a text prompt such as category names or referring expressions on images. The
+    categories in text prompt are separated by commas. It returns a list of bounding
+    boxes with normalized coordinates, label names and associated probability scores.
+
+    Parameters:
+        prompt (str): The prompt to ground to the image.
+        image (np.ndarray): The image to ground the prompt to.
+        fine_tune_id (Optional[str]): If you have a fine-tuned model, you can pass the
+            fine-tuned model ID here to use it.
+
+    Returns:
+        List[Dict[str, Any]]: A list of dictionaries containing the score, label, and
+            bounding box of the detected objects with normalized coordinates between 0
+            and 1 (xmin, ymin, xmax, ymax). xmin and ymin are the coordinates of the
+            top-left and xmax and ymax are the coordinates of the bottom-right of the
+            bounding box.
+
+    Example
+    -------
+        >>> agentic_object_detection("car", image)
+        [
+            {'score': 0.99, 'label': 'car', 'bbox': [0.1, 0.11, 0.35, 0.4]},
+            {'score': 0.98, 'label': 'car', 'bbox': [0.2, 0.21, 0.45, 0.5},
+        ]
+    """
+
+    image_size = image.shape[:2]
+    if image_size[0] < 1 or image_size[1] < 1:
+        return []
+
+    ret = _agentic_object_detection(
+        prompt, image, image_size, fine_tune_id=fine_tune_id
+    )
+
+    _display_tool_trace(
+        agentic_object_detection.__name__,
+        {"prompts": prompt},
+        ret["display_data"],
+        ret["files"],
+    )
+    return ret["return_data"]  # type: ignore
+
+
+def agentic_sam2_instance_segmentation(
+    prompt: str, image: np.ndarray
+) -> List[Dict[str, Any]]:
+    """'agentic_sam2_instance_segmentation' is a tool that can detect and count multiple
+    instances of objects given a text prompt such as category names or referring
+    expressions on images. The categories in text prompt are separated by commas. It
+    returns a list of bounding boxes with normalized coordinates, label names, masks
+    and associated probability scores.
+
+    Parameters:
+        prompt (str): The object that needs to be counted.
+        image (np.ndarray): The image that contains multiple instances of the object.
+
+    Returns:
+        List[Dict[str, Any]]: A list of dictionaries containing the score, label,
+            bounding box, and mask of the detected objects with normalized coordinates
+            (xmin, ymin, xmax, ymax). xmin and ymin are the coordinates of the top-left
+            and xmax and ymax are the coordinates of the bottom-right of the bounding box.
+            The mask is binary 2D numpy array where 1 indicates the object and 0 indicates
+            the background.
+
+    Example
+    -------
+        >>> agentic_sam2_instance_segmentation("flower", image)
+        [
+            {
+                'score': 0.49,
+                'label': 'flower',
+                'bbox': [0.1, 0.11, 0.35, 0.4],
+                'mask': array([[0, 0, 0, ..., 0, 0, 0],
+                    [0, 0, 0, ..., 0, 0, 0],
+                    ...,
+                    [0, 0, 0, ..., 0, 0, 0],
+                    [0, 0, 0, ..., 0, 0, 0]], dtype=uint8),
+            },
+        ]
+    """
+
+    od_ret = _agentic_object_detection(prompt, image, image.shape[:2])
+    seg_ret = _sam2(
+        image, od_ret["return_data"], image.shape[:2], image_bytes=od_ret["files"][0][1]
+    )
+
+    _display_tool_trace(
+        agentic_sam2_instance_segmentation.__name__,
+        {
+            "prompts": prompt,
+        },
+        seg_ret["display_data"],
+        seg_ret["files"],
+    )
+
+    return seg_ret["return_data"]  # type: ignore
+
+
+def agentic_sam2_video_tracking(
+    prompt: str,
+    frames: List[np.ndarray],
+    chunk_length: Optional[int] = 10,
+    fine_tune_id: Optional[str] = None,
+) -> List[List[Dict[str, Any]]]:
+    """'agentic_sam2_video_tracking' is a tool that can track and segment multiple
+    objects in a video given a text prompt such as category names or referring
+    expressions. The categories in the text prompt are separated by commas. It returns
+    a list of bounding boxes, label names, masks and associated probability scores and
+    is useful for tracking and counting without duplicating counts.
+
+    Parameters:
+        prompt (str): The prompt to ground to the image.
+        frames (List[np.ndarray]): The list of frames to ground the prompt to.
+        chunk_length (Optional[int]): The number of frames to re-run agentic object detection to
+            to find new objects.
+        fine_tune_id (Optional[str]): If you have a fine-tuned model, you can pass the
+            fine-tuned model ID here to use it.
+
+    Returns:
+        List[List[Dict[str, Any]]]: A list of list of dictionaries containing the
+            label, segmentation mask and bounding boxes. The outer list represents each
+            frame and the inner list is the entities per frame. The detected objects
+            have normalized coordinates between 0 and 1 (xmin, ymin, xmax, ymax). xmin
+            and ymin are the coordinates of the top-left and xmax and ymax are the
+            coordinates of the bottom-right of the bounding box. The mask is binary 2D
+            numpy array where 1 indicates the object and 0 indicates the background.
+            The label names are prefixed with their ID represent the total count.
+
+    Example
+    -------
+        >>> agentic_sam2_video_tracking("dinosaur", frames)
+        [
+            [
+                {
+                    'label': '0: dinosaur',
+                    'bbox': [0.1, 0.11, 0.35, 0.4],
+                    'mask': array([[0, 0, 0, ..., 0, 0, 0],
+                        [0, 0, 0, ..., 0, 0, 0],
+                        ...,
+                        [0, 0, 0, ..., 0, 0, 0],
+                        [0, 0, 0, ..., 0, 0, 0]], dtype=uint8),
+                },
+            ],
+            ...
+        ]
+    """
+
+    ret = od_sam2_video_tracking(
+        ODModels.AGENTIC,
+        prompt=prompt,
+        frames=frames,
+        chunk_length=chunk_length,
+        fine_tune_id=fine_tune_id,
+    )
+    _display_tool_trace(
+        agentic_sam2_video_tracking.__name__,
+        {},
+        ret["display_data"],
+        ret["files"],
+    )
+    return ret["return_data"]  # type: ignore
+
+
 def minimum_distance(
     det1: Dict[str, Any], det2: Dict[str, Any], image_size: Tuple[int, int]
 ) -> float:

{vision_agent-0.2.226 → vision_agent-0.2.228}/vision_agent/utils/video_tracking.py

@@ -17,6 +17,8 @@ class ODModels(str, Enum):
     COUNTGD = "countgd"
     FLORENCE2 = "florence2"
     OWLV2 = "owlv2"
+    AGENTIC = "agentic"
+    CUSTOM = "custom"
 
 
 def split_frames_into_segments(