vision-agent 1.1.16__py3-none-any.whl → 1.1.17__py3-none-any.whl

vision_agent/.sim_tools/df.csv

@@ -388,8 +388,8 @@ desc,doc,name
     -------
         >>> document_qa(image, question)
         'The answer to the question ...'",document_qa
-"'ocr' extracts text from an image. It returns a list of detected text, bounding boxes with normalized coordinates, and confidence scores. The results are sorted from top-left to bottom right.","ocr(image: numpy.ndarray) -> List[Dict[str, Any]]:
-'ocr' extracts text from an image. It returns a list of detected text, bounding
+"'paddle_ocr' extracts text from an image. It returns a list of detected text, bounding boxes with normalized coordinates, and confidence scores. The results are sorted from top-left to bottom right.","paddle_ocr(image: numpy.ndarray) -> List[Dict[str, Any]]:
+'paddle_ocr' extracts text from an image. It returns a list of detected text, bounding
     boxes with normalized coordinates, and confidence scores. The results are sorted
     from top-left to bottom right.
 
@@ -402,10 +402,10 @@ desc,doc,name
 
     Example
     -------
-        >>> ocr(image)
+        >>> paddle_ocr(image)
         [
             {'label': 'hello world', 'bbox': [0.1, 0.11, 0.35, 0.4], 'score': 0.99},
-        ]",ocr
+        ]",paddle_ocr
 "'gemini_image_generation' performs either image inpainting given an image and text prompt, or image generation given a prompt. It can be used to edit parts of an image or the entire image according to the prompt given.","gemini_image_generation(prompt: str, image: Optional[numpy.ndarray] = None) -> numpy.ndarray:
 'gemini_image_generation' performs either image inpainting given an image and text prompt, or image generation given a prompt.
     It can be used to edit parts of an image or the entire image according to the prompt given.
@@ -484,26 +484,26 @@ desc,doc,name
             {'start_time': 2, 'end_time': 4, 'location': 'Outdoor area', 'description': 'A person approaches a white bicycle parked in a row. The person then swings their leg over the bike and gets on it.', 'label': 0},
             {'start_time': 10, 'end_time': 13, 'location': 'Outdoor area', 'description': 'A person gets off a white bicycle parked in a row. The person swings their leg over the bike and dismounts.', 'label': 1},
         ]",agentic_activity_recognition
-'depth_anything_v2' is a tool that runs depth anything v2 model to generate a depth image from a given RGB image. The returned depth image is monochrome and represents depth values as pixel intensities with pixel values ranging from 0 to 255.,"depth_anything_v2(image: numpy.ndarray) -> numpy.ndarray:
-'depth_anything_v2' is a tool that runs depth anything v2 model to generate a
-    depth image from a given RGB image. The returned depth image is monochrome and
-    represents depth values as pixel intensities with pixel values ranging from 0 to 255.
+"'depth_pro' is a tool that runs the Apple DepthPro model to generate a depth map from a given RGB image. The returned depth map has the same dimensions as the input image, with each pixel indicating the distance from the camera in meters.","depth_pro(image: numpy.ndarray) -> numpy.ndarray:
+'depth_pro' is a tool that runs the Apple DepthPro model to generate a
+    depth map from a given RGB image. The returned depth map has the same dimensions 
+    as the input image, with each pixel indicating the distance from the camera in meters.
 
     Parameters:
         image (np.ndarray): The image to used to generate depth image
 
     Returns:
-        np.ndarray: A grayscale depth image with pixel values ranging from 0 to 255
-            where high values represent closer objects and low values further.
+        np.ndarray: A depth map with float32 pixel values that represent
+            the distance from the camera in meters.
 
     Example
     -------
-        >>> depth_anything_v2(image)
+        >>> depth_pro(image)
         array([[0, 0, 0, ..., 0, 0, 0],
                 [0, 20, 24, ..., 0, 100, 103],
                 ...,
                 [10, 11, 15, ..., 202, 202, 205],
-                [10, 10, 10, ..., 200, 200, 200]], dtype=uint8),",depth_anything_v2
+                [10, 10, 10, ..., 200, 200, 200]], dtype=np.float32),",depth_pro
 'generate_pose_image' is a tool that generates a open pose bone/stick image from a given RGB image. The returned bone image is RGB with the pose amd keypoints colored and background as black.,"generate_pose_image(image: numpy.ndarray) -> numpy.ndarray:
 'generate_pose_image' is a tool that generates a open pose bone/stick image from
     a given RGB image. The returned bone image is RGB with the pose amd keypoints colored

vision_agent/tools/__init__.py

@@ -21,7 +21,7 @@ from .tools import (
     countgd_sam2_visual_instance_segmentation,
     countgd_visual_object_detection,
     custom_object_detection,
-    depth_anything_v2,
+    depth_pro,
     detr_segmentation,
     document_extraction,
     document_qa,
@@ -42,7 +42,7 @@ from .tools import (
     glee_sam2_video_tracking,
     load_image,
     minimum_distance,
-    ocr,
+    paddle_ocr,
     od_sam2_video_tracking,
     overlay_bounding_boxes,
     overlay_heat_map,

vision_agent/tools/tools.py

@@ -4,7 +4,7 @@ import logging
 import os
 import tempfile
 import urllib.request
-from base64 import b64encode
+from base64 import b64encode, b64decode
 from concurrent.futures import ThreadPoolExecutor, as_completed
 from importlib import resources
 from pathlib import Path
@@ -15,7 +15,6 @@ import time
 import cv2
 import numpy as np
 import pandas as pd
-import requests
 from IPython.display import display
 from PIL import Image, ImageDraw, ImageFont
 from pillow_heif import register_heif_opener  # type: ignore
@@ -2034,8 +2033,8 @@ def qwen2_vl_video_vqa(prompt: str, frames: List[np.ndarray]) -> str:
     return cast(str, data)
 
 
-def ocr(image: np.ndarray) -> List[Dict[str, Any]]:
-    """'ocr' extracts text from an image. It returns a list of detected text, bounding
+def paddle_ocr(image: np.ndarray) -> List[Dict[str, Any]]:
+    """'paddle_ocr' extracts text from an image. It returns a list of detected text, bounding
     boxes with normalized coordinates, and confidence scores. The results are sorted
     from top-left to bottom right.
 
@@ -2048,51 +2047,33 @@ def ocr(image: np.ndarray) -> List[Dict[str, Any]]:
 
     Example
     -------
-        >>> ocr(image)
+        >>> paddle_ocr(image)
         [
             {'label': 'hello world', 'bbox': [0.1, 0.11, 0.35, 0.4], 'score': 0.99},
         ]
     """
 
-    pil_image = Image.fromarray(image).convert("RGB")
-    image_size = pil_image.size[::-1]
+    image_size = image.shape[:2]
     if image_size[0] < 1 or image_size[1] < 1:
         return []
-    image_buffer = io.BytesIO()
-    pil_image.save(image_buffer, format="PNG")
-    buffer_bytes = image_buffer.getvalue()
-    image_buffer.close()
-
-    res = requests.post(
-        _OCR_URL,
-        files={"images": buffer_bytes},
-        data={"language": "en"},
-        headers={"contentType": "multipart/form-data", "apikey": _API_KEY},
-    )
-
-    if res.status_code != 200:
-        raise ValueError(f"OCR request failed with status code {res.status_code}")
-
-    data = res.json()
-    output = []
-    for det in data[0]:
-        label = det["text"]
-        box = [
-            det["location"][0]["x"],
-            det["location"][0]["y"],
-            det["location"][2]["x"],
-            det["location"][2]["y"],
-        ]
-        box = normalize_bbox(box, image_size)
-        output.append({"label": label, "bbox": box, "score": round(det["score"], 2)})
+    buffer_bytes = numpy_to_bytes(image)
+    files = [("image", buffer_bytes)]
+
+    res = send_inference_request(
+        payload={"function_name": "paddle-ocr"},
+        endpoint_name="paddle-ocr",
+        files=files,
+        v2=True,
+    )
 
     _display_tool_trace(
-        ocr.__name__,
+        paddle_ocr.__name__,
         {},
-        data,
-        cast(List[Tuple[str, bytes]], [("image", buffer_bytes)]),
+        res,
+        files,
     )
-    return sorted(output, key=lambda x: (x["bbox"][1], x["bbox"][0]))
+
+    return sorted(res, key=lambda x: (x["bbox"][1], x["bbox"][0]))
 
 
 def claude35_text_extraction(image: np.ndarray) -> str:
@@ -2370,7 +2351,12 @@ def agentic_activity_recognition(
     buffer_bytes = frames_to_bytes(frames, fps=fps)
     files = [("video", buffer_bytes)]
 
-    payload = {"prompt": prompt, "specificity": specificity, "with_audio": with_audio}
+    payload = {
+        "prompt": prompt,
+        "specificity": specificity,
+        "with_audio": with_audio,
+        "function_name": "agentic_activity_recognition",
+    }
 
     response = send_inference_request(
         payload=payload, endpoint_name="activity-recognition", files=files, v2=True
@@ -2529,48 +2515,53 @@ def detr_segmentation(image: np.ndarray) -> List[Dict[str, Any]]:
     return return_data
 
 
-def depth_anything_v2(image: np.ndarray) -> np.ndarray:
-    """'depth_anything_v2' is a tool that runs depth anything v2 model to generate a
-    depth image from a given RGB image. The returned depth image is monochrome and
-    represents depth values as pixel intensities with pixel values ranging from 0 to 255.
+def depth_pro(
+    image: np.ndarray,
+) -> np.ndarray:
+    """'depth_pro' is a tool that runs the Apple DepthPro model to generate a
+    depth map from a given RGB image. The returned depth map has the same dimensions
+    as the input image, with each pixel indicating the distance from the camera in meters.
 
     Parameters:
         image (np.ndarray): The image to used to generate depth image
 
     Returns:
-        np.ndarray: A grayscale depth image with pixel values ranging from 0 to 255
-            where high values represent closer objects and low values further.
+        np.ndarray: A depth map with float32 pixel values that represent
+            the distance from the camera in meters.
 
     Example
     -------
-        >>> depth_anything_v2(image)
+        >>> depth_pro(image)
         array([[0, 0, 0, ..., 0, 0, 0],
                 [0, 20, 24, ..., 0, 100, 103],
                 ...,
                 [10, 11, 15, ..., 202, 202, 205],
-                [10, 10, 10, ..., 200, 200, 200]], dtype=uint8),
+                [10, 10, 10, ..., 200, 200, 200]], dtype=np.float32),
     """
-    if image.shape[0] < 1 or image.shape[1] < 1:
-        raise ValueError(f"Image is empty, image shape: {image.shape}")
 
-    image_b64 = convert_to_b64(image)
-    data = {
-        "image": image_b64,
-        "function_name": "depth_anything_v2",
-    }
+    image_size = image.shape[:2]
+    if image_size[0] < 1 or image_size[1] < 1:
+        return np.empty(0)
+    buffer_bytes = numpy_to_bytes(image)
+    files = [("image", buffer_bytes)]
 
-    depth_map = send_inference_request(data, "depth-anything-v2", v2=True)
-    depth_map_np = np.array(depth_map["map"])
-    depth_map_np = (depth_map_np - depth_map_np.min()) / (
-        depth_map_np.max() - depth_map_np.min()
+    detections = send_inference_request(
+        payload={"function_name": "depth-pro"},
+        endpoint_name="depth-pro",
+        files=files,
+        v2=True,
     )
-    depth_map_np = (255 * depth_map_np).astype(np.uint8)
+
+    depth_bytes = b64decode(detections["depth"])
+    depth_map_np = np.frombuffer(depth_bytes, dtype=np.float32).reshape(image_size)
+
     _display_tool_trace(
-        depth_anything_v2.__name__,
+        depth_pro.__name__,
         {},
-        depth_map,
-        image_b64,
+        response=detections,
+        files=files,
     )
+
     return depth_map_np
 
 
@@ -3564,12 +3555,12 @@ FUNCTION_TOOLS = [
     claude35_text_extraction,
     agentic_document_extraction,
     document_qa,
-    ocr,
+    paddle_ocr,
     gemini_image_generation,
     qwen25_vl_images_vqa,
     qwen25_vl_video_vqa,
     agentic_activity_recognition,
-    depth_anything_v2,
+    depth_pro,
     generate_pose_image,
     vit_nsfw_classification,
     siglip_classification,

{vision_agent-1.1.16.dist-info → vision_agent-1.1.17.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: vision-agent
-Version: 1.1.16
+Version: 1.1.17
 Summary: Toolset for Vision Agent
 Project-URL: Homepage, https://landing.ai
 Project-URL: repository, https://github.com/landing-ai/vision-agent

{vision_agent-1.1.16.dist-info → vision_agent-1.1.17.dist-info}/RECORD

@@ -1,6 +1,6 @@
 vision_agent/__init__.py,sha256=EAb4-f9iyuEYkBrX4ag1syM8Syx8118_t0R6_C34M9w,57
-vision_agent/.sim_tools/df.csv,sha256=i732_U1KQf55UNhT-9srtZXF91XvDnfWBDdc8EqDmpw,41215
-vision_agent/.sim_tools/embs.npy,sha256=XCu3LnLS10IS3npfPMqX2VHIbDPq9iY_NPDBwq5AEj0,245888
+vision_agent/.sim_tools/df.csv,sha256=gheT5OXu68o0AfjV1623GzbD-T2csZ7GnkBbCMaVl8c,41188
+vision_agent/.sim_tools/embs.npy,sha256=OLj2rt4aBFze2HIf9bQ3yn0-_3RVPecrHWxm2CWvgn0,245888
 vision_agent/agent/README.md,sha256=3XSPG_VO7-6y6P8COvcgSSonWj5uvfgvfmOkBpfKK8Q,5527
 vision_agent/agent/__init__.py,sha256=_-nGLHhRTLViXxBSb9D4OwLTqk9HXKPEkTBkvK8c7OU,206
 vision_agent/agent/agent.py,sha256=o1Zuhl6h2R7uVwvUur0Aj38kak8U08plfeFWPst_ErM,1576
@@ -26,11 +26,11 @@ vision_agent/models/lmm_types.py,sha256=v04h-NjbczHOIN8UWa1vvO5-1BDuZ4JQhD2mge1c
 vision_agent/models/tools_types.py,sha256=8hYf2OZhI58gvf65KGaeGkt4EQ56nwLFqIQDPHioOBc,2339
 vision_agent/sim/__init__.py,sha256=Aouz6HEPPTYcLxR5_0fTYCL1OvPKAH1RMWAF90QXAlA,135
 vision_agent/sim/sim.py,sha256=WQY_x9A4VT647qGDBScJ3R8_Iv0aoYLHTgwcQSCXwv4,10059
-vision_agent/tools/__init__.py,sha256=zf8HzjcMSgxKhtrxbqYe9hmvsfuweeDMrOc8eVA8Ya8,2477
+vision_agent/tools/__init__.py,sha256=WfynKGn0Zl2GPkyFhzA2YhGGC0Dtb1oei4Hk_GdSY1c,2476
 vision_agent/tools/meta_tools.py,sha256=9iJilpGYEiXW0nYPTYAWHa7l23wGN8IM5KbE7mWDOT0,6798
 vision_agent/tools/planner_tools.py,sha256=iQWtTgXdomn0IWrbmvXXM-y8Q_RSEOxyP04HIRLrgWI,19576
 vision_agent/tools/prompts.py,sha256=V1z4YJLXZuUl_iZ5rY0M5hHc_2tmMEUKr0WocXKGt4E,1430
-vision_agent/tools/tools.py,sha256=i9GGGu8tvo2M6O5fF4UUBTpn_Ul2KEN9mG3ZlJ95qao,124929
+vision_agent/tools/tools.py,sha256=lndSG8xrIWcs6Rpe1-Jq44niUDXQnWlYfGP2B1YjpI0,124216
 vision_agent/utils/__init__.py,sha256=mANUs_84VL-3gpZbXryvV2mWU623eWnRlJCSUHtMjuw,122
 vision_agent/utils/agent.py,sha256=2ifTP5QElItnr4YHOJR6L5P1PUzV0GhChTTqVxuVyQg,15153
 vision_agent/utils/exceptions.py,sha256=zis8smCbdEylBVZBTVfEUfAh7Rb7cWV3MSPambu6FsQ,1837
@@ -40,7 +40,7 @@ vision_agent/utils/tools.py,sha256=Days0dETPRQLSDamMKPnXFsc5g5IKX9QJcPPNmSHNdM,8
 vision_agent/utils/tools_doc.py,sha256=PKcXXbJktiuPi9q6Q1zXzFx24Dh229SNgWBDtZ2fQSQ,2730
 vision_agent/utils/video.py,sha256=rjsQ1sKKisaQ6AVjJz0zd_G4g-ovRweS_rs4JEhenoI,5340
 vision_agent/utils/video_tracking.py,sha256=DZLFpNCuzuPJQzbQoVNcp-m4dKxgiKdCNM5QTh_zURE,12245
-vision_agent-1.1.16.dist-info/METADATA,sha256=JMmL6rIdT1-WO6XTrjNHucAp4S_UlkjDW1dxznQJ994,12078
-vision_agent-1.1.16.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-vision_agent-1.1.16.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-vision_agent-1.1.16.dist-info/RECORD,,
+vision_agent-1.1.17.dist-info/METADATA,sha256=LDH3i8vb2g6aqoEuRSPHdigP1bmhBjxZTQ37-cD9RlA,12078
+vision_agent-1.1.17.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+vision_agent-1.1.17.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+vision_agent-1.1.17.dist-info/RECORD,,