vision-agent 0.2.154__py3-none-any.whl → 0.2.156__py3-none-any.whl

vision_agent/agent/agent_utils.py

@@ -77,3 +77,9 @@ def extract_code(code: str) -> str:
     if code.startswith("python\n"):
         code = code[len("python\n") :]
     return code
+
+
+def remove_installs_from_code(code: str) -> str:
+    pattern = r"\n!pip install.*?(\n|\Z)\n"
+    code = re.sub(pattern, "", code, flags=re.DOTALL)
+    return code

vision_agent/agent/vision_agent.py

@@ -407,8 +407,6 @@ class VisionAgent(Agent):
             code_interpreter.download_file(
                 str(remote_artifacts_path.name), str(self.local_artifacts_path)
             )
-            artifacts.load(self.local_artifacts_path)
-            artifacts.save()
         return orig_chat, artifacts
 
     def streaming_message(self, message: Dict[str, Any]) -> None:

vision_agent/agent/vision_agent_coder.py

@@ -13,7 +13,11 @@ from tabulate import tabulate
 
 import vision_agent.tools as T
 from vision_agent.agent import Agent
-from vision_agent.agent.agent_utils import extract_code, extract_json
+from vision_agent.agent.agent_utils import (
+    extract_code,
+    extract_json,
+    remove_installs_from_code,
+)
 from vision_agent.agent.vision_agent_coder_prompts import (
     CODE,
     FIX_BUG,
@@ -836,8 +840,8 @@ class VisionAgentCoder(Agent):
                 media=media_list,
             )
             success = cast(bool, results["success"])
-            code = cast(str, results["code"])
-            test = cast(str, results["test"])
+            code = remove_installs_from_code(cast(str, results["code"]))
+            test = remove_installs_from_code(cast(str, results["test"]))
             working_memory.extend(results["working_memory"])  # type: ignore
             plan.append({"code": code, "test": test, "plan": plan_i})
 

vision_agent/agent/vision_agent_prompts.py

@@ -28,7 +28,8 @@ Here is the current conversation so far:
 1. **Understand and Clarify**: Make sure you understand the task, ask clarifying questions if the task is not clear.
 2. **Code Generation**: Only use code provided in the Documentation in your <execute_python> tags. Only use `edit_vision_code` to modify code written by `generate_vision_code`.
 3. **Execute**: Do only what the user asked you to do and no more. If you need to ask the user a question, set `let_user_respond` to `true`.
-4. **Output in JSON**: Respond in the following format in JSON:
+4. **Response**: Keep your responses short and concise. Provide the user only with the information they need to continue the conversation.
+5. **Output in JSON**: Respond in the following format in JSON:
 
 ```json
 {{"thoughts": <your thoughts>, "response": <your response to the user>, "let_user_respond": <a boolean whether or not to let the user respond>}}.
@@ -62,7 +63,7 @@ OBSERVATION:
 [{'score': 0.99, 'label': 'dog', 'box': [0.1, 0.2, 0.3, 0.4]}, {'score': 0.23, 'label': 'dog', 'box': [0.2, 0.3, 0.4, 0.5]}]
 
 
-AGENT: {"thoughts": "Two dogs are detected, I will show this to the user and ask them if the result looks good.", "response": "I have written the code to detect dogs and shown the output, do the results look good to you?", "let_user_respond": true}
+AGENT: {"thoughts": "Two dogs are detected, I will show this to the user and ask them if the result looks good.", "response": "The code detectd two dogs, do the results look good to you?", "let_user_respond": true}
 """
 
 EXAMPLES_CODE1_EXTRA = """
@@ -91,7 +92,7 @@ OBSERVATION:
 ----- stdout -----
 [{'score': 0.99, 'label': 'dog', 'box': [0.1, 0.2, 0.3, 0.4]}]
 
-AGENT: {"thoughts": "One dog is detected, I will show this to the user and ask them if the result looks good.", "response": "I have written the code to detect one dog and shown you the output, do the results look good to you?", "let_user_respond": true}
+AGENT: {"thoughts": "One dog is detected, I will show this to the user and ask them if the result looks good.", "response": "The code detected one dog, do these results look good to you?", "let_user_respond": true}
 """
 
 EXAMPLES_CODE2 = """
@@ -157,16 +158,16 @@ OBSERVATION:
 ----- stdout -----
 2
 
-AGENT: {"thoughts": "Two workers with helmets are detected, I will show this to the user and ask them if the result looks good.", "response": "I have written the code to count the workers wearing helmets in code.py and saved the visualization under 'workers_viz.png'.", "let_user_respond": true}
+AGENT: {"thoughts": "Two workers with helmets are detected, I will show this to the user and ask them if the result looks good.", "response": "The code to detect workers with helmets is saved in code.py and the visualization under 'workers_viz.png'.", "let_user_respond": true}
 
 USER: The detections are slightly off. Can you fine tune florence2 using these labels? "[{'image_path': 'image1.png': 'labels': ['worker', 'helmet'], 'bboxes': [[235, 118, 294, 241], [232, 118, 294, 128]]}, {'image_path': 'image2.png': 'labels': ['worker', 'helmet'], 'bboxes': [[235, 118, 294, 241], [232, 118, 294, 128]]}]"
 
-AGENT: {"thoughts": "Because the user has supplied me with labels I can call object_detection_fine_tuning on their behalf to fine tune the model", "response": "I will fine tune florence2 with the labels you provided <execute_python>object_detection_fine_tuning([{'image_path': 'image1.png': 'labels': ['worker', 'helmet'], 'bboxes': [[235, 118, 294, 241], [232, 118, 294, 128]]}, {'image_path': 'image2.png': 'labels': ['worker', 'helmet'], 'bboxes': [[235, 118, 294, 241], [232, 118, 294, 128]]}])</execute_python>", "let_user_respond": false}
+AGENT: {"thoughts": "Because the user has supplied me with labels I can call object_detection_fine_tuning on their behalf to fine tune the model", "response": "<execute_python>object_detection_fine_tuning([{'image_path': 'image1.png': 'labels': ['worker', 'helmet'], 'bboxes': [[235, 118, 294, 241], [232, 118, 294, 128]]}, {'image_path': 'image2.png': 'labels': ['worker', 'helmet'], 'bboxes': [[235, 118, 294, 241], [232, 118, 294, 128]]}])</execute_python>", "let_user_respond": false}
 
 OBSERVATION:
 [Fine tuning id: 23b3b022-5ebf-4798-9373-20ef36429abf]
 
-AGENT: {"thoughts": "The model has finished fine tuning, I will now replace the original florence2_phrase_grounding call with the fine tuning id.", "response": "I will not update the code to use the fine tuned model. <execute_python>use_object_detection_fine_tuning(artifacts, "code.py", "23b3b022-5ebf-4798-9373-20ef36429abf")</execute_python>", "let_user_respond": false}
+AGENT: {"thoughts": "The model has finished fine tuning, I will now replace the original florence2_phrase_grounding call with the fine tuning id.", "response": "<execute_python>use_object_detection_fine_tuning(artifacts, "code.py", "23b3b022-5ebf-4798-9373-20ef36429abf")</execute_python>", "let_user_respond": false}
 
 OBSERVATION:
 [Artifact code.py edits]

vision_agent/tools/__init__.py

@@ -45,7 +45,6 @@ from .tools import (
     loca_zero_shot_counting,
     ocr,
     overlay_bounding_boxes,
-    overlay_counting_results,
     overlay_heat_map,
     overlay_segmentation_masks,
     owl_v2_image,

vision_agent/tools/meta_tools.py

@@ -116,7 +116,9 @@ class Artifacts:
         )
         output_str = "[Artifacts loaded]\n"
         for k in self.artifacts.keys():
-            output_str += f"Artifact {k} loaded to {str(loaded_path / k)}\n"
+            output_str += (
+                f"Artifact name: {k}, loaded to path: {str(loaded_path / k)}\n"
+            )
         output_str += "[End of artifacts]\n"
         print(output_str)
         return output_str

vision_agent/tools/tool_utils.py

@@ -1,6 +1,6 @@
-import os
 import inspect
 import logging
+import os
 from base64 import b64encode
 from typing import Any, Callable, Dict, List, MutableMapping, Optional, Tuple
 

vision_agent/tools/tools.py

@@ -13,7 +13,7 @@ from uuid import UUID
 import cv2
 import numpy as np
 import requests
-from PIL import Image, ImageDraw, ImageEnhance, ImageFont
+from PIL import Image, ImageDraw, ImageFont
 from pillow_heif import register_heif_opener  # type: ignore
 from pytube import YouTube  # type: ignore
 
@@ -700,18 +700,22 @@ def countgd_counting(
             {'score': 0.98, 'label': 'flower', 'bbox': [0.44, 0.24, 0.49, 0.58},
         ]
     """
-    image_b64 = convert_to_b64(image)
+    buffer_bytes = numpy_to_bytes(image)
+    files = [("image", buffer_bytes)]
     prompt = prompt.replace(", ", " .")
-    payload = {"prompt": prompt, "image": image_b64}
+    payload = {"prompts": [prompt], "model": "countgd"}
     metadata = {"function_name": "countgd_counting"}
-    resp_data = send_task_inference_request(payload, "countgd", metadata=metadata)
+    resp_data = send_task_inference_request(
+        payload, "text-to-object-detection", files=files, metadata=metadata
+    )
+    bboxes_per_frame = resp_data[0]
     bboxes_formatted = [
         ODResponseData(
             label=bbox["label"],
-            bbox=list(map(lambda x: round(x, 2), bbox["bbox"])),
+            bbox=list(map(lambda x: round(x, 2), bbox["bounding_box"])),
             score=round(bbox["score"], 2),
         )
-        for bbox in resp_data
+        for bbox in bboxes_per_frame
     ]
     filtered_bboxes = filter_bboxes_by_threshold(bboxes_formatted, box_threshold)
     return [bbox.model_dump() for bbox in filtered_bboxes]
@@ -1146,10 +1150,10 @@ def florence2_image_caption(image: np.ndarray, detail_caption: bool = True) -> s
 def florence2_phrase_grounding(
     prompt: str, image: np.ndarray, fine_tune_id: Optional[str] = None
 ) -> List[Dict[str, Any]]:
-    """'florence2_phrase_grounding' will run florence2 on a image. It can
-    detect multiple objects given a text prompt which can be object names or caption.
-    You can optionally separate the object names in the text with commas. It returns
-    a list of bounding boxes with normalized coordinates, label names and associated
+    """'florence2_phrase_grounding' is a tool that can detect multiple
+    objects given a text prompt which can be object names or caption. You
+    can optionally separate the object names in the text with commas. It returns a list
+    of bounding boxes with normalized coordinates, label names and associated
     probability scores of 1.0.
 
     Parameters:
@@ -1808,6 +1812,11 @@ def save_image(image: np.ndarray, file_path: str) -> None:
     """
     from IPython.display import display
 
+    if not isinstance(image, np.ndarray) or (
+        image.shape[0] == 0 and image.shape[1] == 0
+    ):
+        raise ValueError("The image is not a valid NumPy array with shape (H, W, C)")
+
     pil_image = Image.fromarray(image.astype(np.uint8)).convert("RGB")
     display(pil_image)
     pil_image.save(file_path)
@@ -1834,6 +1843,15 @@ def save_video(
     if fps <= 0:
         raise ValueError(f"fps must be greater than 0 got {fps}")
 
+    if not isinstance(frames, list) or len(frames) == 0:
+        raise ValueError("Frames must be a list of NumPy arrays")
+
+    for frame in frames:
+        if not isinstance(frame, np.ndarray) or (
+            frame.shape[0] == 0 and frame.shape[1] == 0
+        ):
+            raise ValueError("A frame is not a valid NumPy array with shape (H, W, C)")
+
     if output_video_path is None:
         output_video_path = tempfile.NamedTemporaryFile(
             delete=False, suffix=".mp4"
@@ -1903,30 +1921,36 @@ def overlay_bounding_boxes(
         bboxes = bbox_int[i]
         bboxes = sorted(bboxes, key=lambda x: x["label"], reverse=True)
 
-        width, height = pil_image.size
-        fontsize = max(12, int(min(width, height) / 40))
-        draw = ImageDraw.Draw(pil_image)
-        font = ImageFont.truetype(
-            str(
-                resources.files("vision_agent.fonts").joinpath("default_font_ch_en.ttf")
-            ),
-            fontsize,
-        )
-
-        for elt in bboxes:
-            label = elt["label"]
-            box = elt["bbox"]
-            scores = elt["score"]
-
-            # denormalize the box if it is normalized
-            box = denormalize_bbox(box, (height, width))
-            draw.rectangle(box, outline=color[label], width=4)
-            text = f"{label}: {scores:.2f}"
-            text_box = draw.textbbox((box[0], box[1]), text=text, font=font)
-            draw.rectangle(
-                (box[0], box[1], text_box[2], text_box[3]), fill=color[label]
+        if len(bboxes) > 20:
+            pil_image = _plot_counting(pil_image, bboxes, color)
+        else:
+            width, height = pil_image.size
+            fontsize = max(12, int(min(width, height) / 40))
+            draw = ImageDraw.Draw(pil_image)
+            font = ImageFont.truetype(
+                str(
+                    resources.files("vision_agent.fonts").joinpath(
+                        "default_font_ch_en.ttf"
+                    )
+                ),
+                fontsize,
             )
-            draw.text((box[0], box[1]), text, fill="black", font=font)
+
+            for elt in bboxes:
+                label = elt["label"]
+                box = elt["bbox"]
+                scores = elt["score"]
+
+                # denormalize the box if it is normalized
+                box = denormalize_bbox(box, (height, width))
+                draw.rectangle(box, outline=color[label], width=4)
+                text = f"{label}: {scores:.2f}"
+                text_box = draw.textbbox((box[0], box[1]), text=text, font=font)
+                draw.rectangle(
+                    (box[0], box[1], text_box[2], text_box[3]), fill=color[label]
+                )
+                draw.text((box[0], box[1]), text, fill="black", font=font)
+
         frame_out.append(np.array(pil_image))
     return frame_out[0] if len(frame_out) == 1 else frame_out
 
@@ -2085,39 +2109,19 @@ def overlay_heat_map(
     return np.array(combined)
 
 
-def overlay_counting_results(
-    image: np.ndarray, instances: List[Dict[str, Any]]
-) -> np.ndarray:
-    """'overlay_counting_results' is a utility function that displays counting results on
-    an image.
-
-    Parameters:
-        image (np.ndarray): The image to display the bounding boxes on.
-        instances (List[Dict[str, Any]]): A list of dictionaries containing the bounding
-            box information of each instance
-
-    Returns:
-        np.ndarray: The image with the instance_id dislpayed
-
-    Example
-    -------
-        >>> image_with_bboxes = overlay_counting_results(
-            image, [{'score': 0.99, 'label': 'object', 'bbox': [0.1, 0.11, 0.35, 0.4]}],
-        )
-    """
-    pil_image = Image.fromarray(image.astype(np.uint8)).convert("RGB")
-    color = (158, 218, 229)
-
-    width, height = pil_image.size
+def _plot_counting(
+    image: Image.Image,
+    bboxes: List[Dict[str, Any]],
+    colors: Dict[str, Tuple[int, int, int]],
+) -> Image.Image:
+    width, height = image.size
     fontsize = max(10, int(min(width, height) / 80))
-    pil_image = ImageEnhance.Brightness(pil_image).enhance(0.5)
-    draw = ImageDraw.Draw(pil_image)
+    draw = ImageDraw.Draw(image)
     font = ImageFont.truetype(
         str(resources.files("vision_agent.fonts").joinpath("default_font_ch_en.ttf")),
         fontsize,
     )
-
-    for i, elt in enumerate(instances, 1):
+    for i, elt in enumerate(bboxes, 1):
         label = f"{i}"
         box = elt["bbox"]
 
@@ -2139,7 +2143,7 @@ def overlay_counting_results(
         text_y1 = cy + text_height / 2
 
         # Draw the rectangle encapsulating the text
-        draw.rectangle((text_x0, text_y0, text_x1, text_y1), fill=color)
+        draw.rectangle((text_x0, text_y0, text_x1, text_y1), fill=colors[elt["label"]])
 
         # Draw the text at the center of the bounding box
         draw.text(
@@ -2150,7 +2154,7 @@ def overlay_counting_results(
             anchor="lt",
         )
 
-    return np.array(pil_image)
+    return image
 
 
 FUNCTION_TOOLS = [
@@ -2183,7 +2187,6 @@ UTIL_TOOLS = [
     overlay_bounding_boxes,
     overlay_segmentation_masks,
     overlay_heat_map,
-    overlay_counting_results,
 ]
 
 TOOLS = FUNCTION_TOOLS + UTIL_TOOLS

vision_agent/tools/tools_types.py

@@ -1,6 +1,6 @@
 from enum import Enum
-from uuid import UUID
 from typing import List, Optional, Tuple, Union
+from uuid import UUID
 
 from pydantic import BaseModel, ConfigDict, Field, SerializationInfo, field_serializer
 

{vision_agent-0.2.154.dist-info → vision_agent-0.2.156.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: vision-agent
-Version: 0.2.154
+Version: 0.2.156
 Summary: Toolset for Vision Agent
 Author: Landing AI
 Author-email: dev@landing.ai

{vision_agent-0.2.154.dist-info → vision_agent-0.2.156.dist-info}/RECORD

@@ -1,11 +1,11 @@
 vision_agent/__init__.py,sha256=EAb4-f9iyuEYkBrX4ag1syM8Syx8118_t0R6_C34M9w,57
 vision_agent/agent/__init__.py,sha256=NF2LABqHixLvbsOIO-fe-VKZ7awvShLtcT0oQT4eWtI,235
 vision_agent/agent/agent.py,sha256=2cjIOxEuSJrqbfPXYoV0qER5ihXsPFCoEFJa4jpqan0,597
-vision_agent/agent/agent_utils.py,sha256=PEUHqvnHmFL4np_TeFmKMwr5s_dWfdfJz6TF_ogd1dU,2353
-vision_agent/agent/vision_agent.py,sha256=m7apb1smJbRyj0VAellrN_mDrSPAee4DVm6FWRa-e78,18459
-vision_agent/agent/vision_agent_coder.py,sha256=9BT4gaXsqH5pvxo8WGwJN9MTvP1V3TgoJHBpjtlKP9I,38417
+vision_agent/agent/agent_utils.py,sha256=eIpLz2NunEqEsBBrECJaD34-2uY0bsFNnW-XKfqqohs,2518
+vision_agent/agent/vision_agent.py,sha256=wrfAWGLcJMJ62ATFLl0E0-2xszi9HQ4Amp82B7-_Ihw,18376
+vision_agent/agent/vision_agent_coder.py,sha256=2ZoGikn2nakGDfs20XRshZjQUyvbw6l47UhExJAYkqI,38515
 vision_agent/agent/vision_agent_coder_prompts.py,sha256=BmbTMhth4v1qLexuoSeyo47QQ0kPQvL1pLbCJHMsWDw,18910
-vision_agent/agent/vision_agent_prompts.py,sha256=3n92aF-jpUyyrAy06izdHIMPEMZPKD1JV0wfQvt-PD8,11251
+vision_agent/agent/vision_agent_prompts.py,sha256=LZ9Bnx7ZFkqbNOMqwfdiWZU4niND9Z1ArcFHNSn_jzA,11187
 vision_agent/clients/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 vision_agent/clients/http.py,sha256=k883i6M_4nl7zwwHSI-yP5sAgQZIDPM1nrKD6YFJ3Xs,2009
 vision_agent/clients/landing_public_api.py,sha256=lU2ev6E8NICmR8DMUljuGcVFy5VNJQ4WQkWC8WnnJEc,1503
@@ -14,12 +14,12 @@ vision_agent/fonts/default_font_ch_en.ttf,sha256=1YM0Z3XqLDjSNbF7ihQFSAIUdjF9m1r
 vision_agent/lmm/__init__.py,sha256=jyY1sJb_tYKg5-Wzs3p1lvwFkc-aUNZfMcLy3TOC4Zg,100
 vision_agent/lmm/lmm.py,sha256=B5ClgwvbybVCWkf9opDMLjTtJZemUU4KUkQoRxGh43I,16787
 vision_agent/lmm/types.py,sha256=ZEXR_ptBL0ZwDMTDYkgxUCmSZFmBYPQd2jreNzr_8UY,221
-vision_agent/tools/__init__.py,sha256=u41fm9KGX1s9DWzVAGnuungEooxH4X8fSDk5hjXvDiY,2450
-vision_agent/tools/meta_tools.py,sha256=FN2oMhXzCzSzmk6Na6uKw1r5-CGO3lCk94izcWNFKwA,25167
+vision_agent/tools/__init__.py,sha256=PLVbfTMjKxQlHIRWnq9b785W9a52AXQS_tOa0tkQ0ZY,2420
+vision_agent/tools/meta_tools.py,sha256=Xu5h92YRfsbvW_iivTnOhlNAPOc2z7CShjOz8KLI4KA,25212
 vision_agent/tools/prompts.py,sha256=V1z4YJLXZuUl_iZ5rY0M5hHc_2tmMEUKr0WocXKGt4E,1430
-vision_agent/tools/tool_utils.py,sha256=FTSboDmYPQLmIhsc9FeydcrdRZU6huBZKnyBmm0VsHE,8196
-vision_agent/tools/tools.py,sha256=l6PreDTod2tJYuaaWxJljJ7PDsKtBV3YJ-nHgl0AzeI,78332
-vision_agent/tools/tools_types.py,sha256=Qijj5NmY6_Aq1fYwuQYf3J1TAQYTz_1mWkX3Dq4d4e0,2339
+vision_agent/tools/tool_utils.py,sha256=VPGqGJ2ZYEJA6AW7K9X7hQv6vRlMtAQcybE4izdToCw,8196
+vision_agent/tools/tools.py,sha256=aP4GCeuGJDMQAIajflgKPVMjrs7ecdEuNiA9GDnV-Pk,78470
+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
 vision_agent/utils/execute.py,sha256=FqSOr5gtBeKB1g2hbV6-bhox6qItDQNn2o9efq1w6f4,28017
@@ -27,7 +27,7 @@ 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=xbMEoRk13l4fHeQlbvMQhLCn8RNndYmsDhUf01TUeR8,4781
-vision_agent-0.2.154.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-vision_agent-0.2.154.dist-info/METADATA,sha256=gpOhgbgqKO1ypr5HfV4ZLpn0jFA3_v5MX4qNe08154g,13758
-vision_agent-0.2.154.dist-info/WHEEL,sha256=7Z8_27uaHI_UZAc4Uox4PpBhQ9Y5_modZXWMxtUi4NU,88
-vision_agent-0.2.154.dist-info/RECORD,,
+vision_agent-0.2.156.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+vision_agent-0.2.156.dist-info/METADATA,sha256=sVOUUcMWZhi5qijAnzT7VrFUFZuIHKqRqU0cdl9eDoc,13758
+vision_agent-0.2.156.dist-info/WHEEL,sha256=7Z8_27uaHI_UZAc4Uox4PpBhQ9Y5_modZXWMxtUi4NU,88
+vision_agent-0.2.156.dist-info/RECORD,,