vision-agent 0.2.184__py3-none-any.whl → 0.2.186__py3-none-any.whl

vision_agent/tools/tools.py

@@ -28,10 +28,8 @@ from vision_agent.tools.tool_utils import (
     send_task_inference_request,
 )
 from vision_agent.tools.tools_types import (
-    Florence2FtRequest,
     JobStatus,
     ODResponseData,
-    PromptTask,
 )
 from vision_agent.utils.exceptions import FineTuneModelIsNotReady
 from vision_agent.utils.execute import FileSerializer, MimeType
@@ -421,8 +419,15 @@ def florence2_sam2_image(
     if image.shape[0] < 1 or image.shape[1] < 1:
         return []
 
+    buffer_bytes = numpy_to_bytes(image)
+    files = [("image", buffer_bytes)]
+    payload = {
+        "prompt": prompt,
+        "model": "florence2sam2",
+    }
+    metadata = {"function_name": "florence2_sam2_image"}
+
     if fine_tune_id is not None:
-        image_b64 = convert_to_b64(image)
         landing_api = LandingPublicAPI()
         status = landing_api.check_fine_tuning_job(UUID(fine_tune_id))
         if status is not JobStatus.SUCCEEDED:
@@ -430,58 +435,31 @@ def florence2_sam2_image(
                 f"Fine-tuned model {fine_tune_id} is not ready yet"
             )
 
-        req_data_obj = Florence2FtRequest(
-            image=image_b64,
-            task=PromptTask.PHRASE_GROUNDING,
-            prompt=prompt,
-            postprocessing="sam2",
-            job_id=UUID(fine_tune_id),
-        )
-        req_data = req_data_obj.model_dump(by_alias=True, exclude_none=True)
-        detections_ft = send_inference_request(
-            req_data,
-            "florence2-ft",
-            v2=True,
-            is_form=True,
-            metadata_payload={"function_name": "florence2_sam2_image"},
-        )
-        # get the first frame
-        detection = detections_ft[0]
-        return_data = []
-        for i in range(len(detection["bboxes"])):
-            return_data.append(
-                {
-                    "score": 1.0,
-                    "label": detection["labels"][i],
-                    "bbox": normalize_bbox(
-                        detection["bboxes"][i], detection["masks"][i]["size"]
-                    ),
-                    "mask": rle_decode_array(detection["masks"][i]),
-                }
-            )
-        return return_data
+        payload["jobId"] = fine_tune_id
 
-    buffer_bytes = numpy_to_bytes(image)
-    files = [("image", buffer_bytes)]
-    payload = {
-        "prompts": [s.strip() for s in prompt.split(",")],
-        "function_name": "florence2_sam2_image",
-    }
-    detections: Dict[str, Any] = send_inference_request(
-        payload, "florence2-sam2", files=files, v2=True
+    detections = send_task_inference_request(
+        payload,
+        "text-to-instance-segmentation",
+        files=files,
+        metadata=metadata,
     )
 
+    # get the first frame
+    frame = detections[0]
     return_data = []
-    for _, data_i in detections["0"].items():
-        mask = rle_decode_array(data_i["mask"])
-        label = data_i["label"]
-        bbox = normalize_bbox(data_i["bounding_box"], data_i["mask"]["size"])
+    for detection in frame:
+        mask = rle_decode_array(detection["mask"])
+        label = detection["label"]
+        bbox = normalize_bbox(detection["bounding_box"], detection["mask"]["size"])
         return_data.append({"label": label, "bbox": bbox, "mask": mask, "score": 1.0})
     return return_data
 
 
 def florence2_sam2_video_tracking(
-    prompt: str, frames: List[np.ndarray], chunk_length: Optional[int] = 3
+    prompt: str,
+    frames: List[np.ndarray],
+    chunk_length: Optional[int] = 3,
+    fine_tune_id: Optional[str] = None,
 ) -> List[List[Dict[str, Any]]]:
     """'florence2_sam2_video_tracking' is a tool that can segment and track multiple
     entities in a video given a text prompt such as category names or referring
@@ -494,6 +472,8 @@ def florence2_sam2_video_tracking(
         frames (List[np.ndarray]): The list of frames to ground the prompt to.
         chunk_length (Optional[int]): The number of frames to re-run florence2 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
@@ -519,24 +499,43 @@ def florence2_sam2_video_tracking(
             ...
         ]
     """
+    if len(frames) == 0:
+        raise ValueError("No frames provided")
 
     buffer_bytes = frames_to_bytes(frames)
     files = [("video", buffer_bytes)]
     payload = {
-        "prompts": [s.strip() for s in prompt.split(",")],
-        "function_name": "florence2_sam2_video_tracking",
+        "prompt": prompt,
+        "model": "florence2sam2",
     }
+    metadata = {"function_name": "florence2_sam2_video_tracking"}
+
     if chunk_length is not None:
-        payload["chunk_length"] = chunk_length  # type: ignore
-    data: Dict[str, Any] = send_inference_request(
-        payload, "florence2-sam2", files=files, v2=True
+        payload["chunk_length_frames"] = chunk_length  # type: ignore
+
+    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"
+            )
+
+        payload["jobId"] = fine_tune_id
+
+    detections = send_task_inference_request(
+        payload,
+        "text-to-instance-segmentation",
+        files=files,
+        metadata=metadata,
     )
+
     return_data = []
-    for frame_i in data.keys():
+    for frame in detections:
         return_frame_data = []
-        for obj_id, data_j in data[frame_i].items():
-            mask = rle_decode_array(data_j["mask"])
-            label = obj_id + ": " + data_j["label"]
+        for detection in frame:
+            mask = rle_decode_array(detection["mask"])
+            label = str(detection["id"]) + ": " + detection["label"]
             return_frame_data.append({"label": label, "mask": mask, "score": 1.0})
         return_data.append(return_frame_data)
     return return_data
@@ -552,7 +551,7 @@ def ocr(image: np.ndarray) -> List[Dict[str, Any]]:
 
     Returns:
         List[Dict[str, Any]]: A list of dictionaries containing the detected text, bbox
-            with nornmalized coordinates, and confidence score.
+            with normalized coordinates, and confidence score.
 
     Example
     -------
@@ -608,7 +607,7 @@ def loca_zero_shot_counting(image: np.ndarray) -> Dict[str, Any]:
 
     Returns:
         Dict[str, Any]: A dictionary containing the key 'count' and the count as a
-            value, e.g. {count: 12} and a heat map for visaulization purposes.
+            value, e.g. {count: 12} and a heat map for visualization purposes.
 
     Example
     -------
@@ -647,7 +646,7 @@ def loca_visual_prompt_counting(
 
     Returns:
         Dict[str, Any]: A dictionary containing the key 'count' and the count as a
-            value, e.g. {count: 12} and a heat map for visaulization purposes.
+            value, e.g. {count: 12} and a heat map for visualization purposes.
 
     Example
     -------
@@ -1058,23 +1057,25 @@ def video_temporal_localization(
     prompt: str,
     frames: List[np.ndarray],
     model: str = "qwen2vl",
-    chunk_length: Optional[float] = None,
-    chunk_length_seconds: Optional[float] = None,
     chunk_length_frames: Optional[int] = 2,
 ) -> List[float]:
-    """'video_temporal_localization' is a tool that can find objects in a video given a question about it.
-    It returns a list of floats with a value of 1.0 if the object to be found is present in the chunk of video being analyzed.
+    """'video_temporal_localization' will run qwen2vl on each chunk_length_frames
+    value selected for the video. It can detect multiple objects independently per
+    chunk_length_frames given a text prompt such as a referring expression
+    but does not track objects across frames.
+    It returns a list of floats with a value of 1.0 if the objects are found in a given
+    chunk_length_frames of the video.
 
     Parameters:
         prompt (str): The question about the video
         frames (List[np.ndarray]): The reference frames used for the question
-        model (str): The model to use for the inference. Valid values are 'qwen2vl', 'gpt4o', 'internlm-xcomposer'
-        chunk_length (Optional[float]): length of each chunk in seconds
-        chunk_length_seconds (Optional[float]): alternative length for chunk in seconds
+        model (str): The model to use for the inference. Valid values are
+            'qwen2vl', 'gpt4o', 'internlm-xcomposer'
         chunk_length_frames (Optional[int]): length of each chunk in frames
 
     Returns:
-        List[float]: A list of floats with a value of 1.0 if the object to be found is present in the chunk of video
+        List[float]: A list of floats with a value of 1.0 if the objects to be found
+            are present in the chunk_length_frames of the video.
 
     Example
     -------
@@ -1089,10 +1090,6 @@ def video_temporal_localization(
         "model": model,
         "function_name": "video_temporal_localization",
     }
-    if chunk_length is not None:
-        payload["chunk_length"] = chunk_length
-    if chunk_length_seconds is not None:
-        payload["chunk_length_seconds"] = chunk_length_seconds
     if chunk_length_frames is not None:
         payload["chunk_length_frames"] = chunk_length_frames
 
@@ -1791,9 +1788,8 @@ def flux_image_inpainting(
             where 1 indicates areas to be inpainted and 0 indicates areas to be preserved.
 
     Returns:
-        np.ndarray:
-            The generated image(s) as a numpy array in RGB format
-            with values ranging from 0 to 255.
+        np.ndarray: The generated image(s) as a numpy array in RGB format with values
+            ranging from 0 to 255.
 
     -------
     Example:
@@ -2352,6 +2348,7 @@ FUNCTION_TOOLS = [
     closest_box_distance,
     qwen2_vl_images_vqa,
     qwen2_vl_video_vqa,
+    video_temporal_localization,
 ]
 
 UTIL_TOOLS = [

{vision_agent-0.2.184.dist-info → vision_agent-0.2.186.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: vision-agent
-Version: 0.2.184
+Version: 0.2.186
 Summary: Toolset for Vision Agent
 Author: Landing AI
 Author-email: dev@landing.ai
@@ -63,10 +63,10 @@ code to solve the task for them. Check out our discord for updates and roadmaps!
 ## Table of Contents
 - [🚀Quick Start](#quick-start)
 - [📚Documentation](#documentation)
-- [🔍🤖VisionAgent](#vision-agent-basic-usage)
+- [🔍🤖VisionAgent](#visionagent-basic-usage)
 - [🛠️Tools](#tools)
 - [🤖LMMs](#lmms)
-- [💻🤖VisionAgent Coder](#vision-agent-coder)
+- [💻🤖VisionAgent Coder](#visionagent-coder)
 - [🏗️Additional Backends](#additional-backends)
 
 ## Quick Start

{vision_agent-0.2.184.dist-info → vision_agent-0.2.186.dist-info}/RECORD

@@ -20,7 +20,7 @@ vision_agent/tools/__init__.py,sha256=KVP4_6qxOb2lpFdQgQtyDfdkMLL1O6wVZNK19MXp-x
 vision_agent/tools/meta_tools.py,sha256=by7TIbH7lsLIayX_Pe2mS1iw8aeLn2T8yqAo8SkB9Kg,32074
 vision_agent/tools/prompts.py,sha256=V1z4YJLXZuUl_iZ5rY0M5hHc_2tmMEUKr0WocXKGt4E,1430
 vision_agent/tools/tool_utils.py,sha256=VPGqGJ2ZYEJA6AW7K9X7hQv6vRlMtAQcybE4izdToCw,8196
-vision_agent/tools/tools.py,sha256=kHeBjiVvncQJeL_Gni84bgHOCgxko4XO7otpt8IyWU4,83610
+vision_agent/tools/tools.py,sha256=-oq8jzITi-yVYJ3ut5MuGJ65jd3ESRtHfw4SCAruMps,83059
 vision_agent/tools/tools_types.py,sha256=8hYf2OZhI58gvf65KGaeGkt4EQ56nwLFqIQDPHioOBc,2339
 vision_agent/utils/__init__.py,sha256=7fMgbZiEwbNS0fBOS_hJI5PuEYBblw36zLi_UjUzvj4,244
 vision_agent/utils/exceptions.py,sha256=booSPSuoULF7OXRr_YbC4dtKt6gM_HyiFQHBuaW86C4,2052
@@ -29,7 +29,7 @@ vision_agent/utils/image_utils.py,sha256=rm9GfXvD4JrjnqKrP_f2gfq4SzmqYC0IdC1kKwd
 vision_agent/utils/sim.py,sha256=ZuSS07TUXFGjipmiQoY8TKRmSes7XXCdtU9PI8PC1sw,5609
 vision_agent/utils/type_defs.py,sha256=BE12s3JNQy36QvauXHjwyeffVh5enfcvd4vTzSwvEZI,1384
 vision_agent/utils/video.py,sha256=fOPR48-SuwMbE5eB5rc2F7lVo6k1mVHn26eEJ0QCslc,5602
-vision_agent-0.2.184.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-vision_agent-0.2.184.dist-info/METADATA,sha256=n8BeCLsPCBXDsr0FCmRBtScseMyJ8TuR68MWlqeO9Is,18330
-vision_agent-0.2.184.dist-info/WHEEL,sha256=7Z8_27uaHI_UZAc4Uox4PpBhQ9Y5_modZXWMxtUi4NU,88
-vision_agent-0.2.184.dist-info/RECORD,,
+vision_agent-0.2.186.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+vision_agent-0.2.186.dist-info/METADATA,sha256=NQfESIRsq9-QWyPzNkyv6dSuRS6TGe5D2tZH4iJpeBU,18328
+vision_agent-0.2.186.dist-info/WHEEL,sha256=7Z8_27uaHI_UZAc4Uox4PpBhQ9Y5_modZXWMxtUi4NU,88
+vision_agent-0.2.186.dist-info/RECORD,,