Tool Updates (#352)

* use hungarian method for matching on tracking

* fix issue where it puts in wrong imports

* update docs

* add activity recognition'

* updated agentic od tools

* updated prompts for agentic od

* updated prompts for agentic od

* update prompts

* update tool retrieval method

* update retrieval prompt

* add back in florence2 tool

* test suggester as o1

* use countgd as example instead of florence2

* do not remove any chats

* add back in florence2 tool

* remove duplicate tools

* handle multiple file types

* resize video so we don't run into issues with size

* update types

* udpate activity tracking

* limit frames, ask for more detailed description for tool choice

* ask to test out several prompts

* updated docs

* updating for hil video

* re-organized to reduce chances of circular imports'

* re-organized to reduce chances of circular imports

* add new judge agent

* re-organized to reduce chances of circular imports

* work for o3

* fix bug with utility docs

* reset judges to 3

* undo agentic OD and judge

* update embs

* add video writing

* fix bug with tools for hil

* add support for o3

* isort

* fix circular import

* add scores to tracking

* display video preds correctly

* updated local app

* fixed import bug

* update example app with new modules

* fix side cases with prompts

* beef up suggester module

* fix type issues

* flake8

* fix new module names

* fix import

* run black

* doc fix from flake8 fix

* update readme

* remove old index

* fix visualization for vqa tools

* addressing github comments

README.md

@@ -40,7 +40,7 @@ You can run VisionAgent in a local Jupyter Notebook [Counting cans in an image](
 You can use VisionAgent to generate code to count the number of people in an image:
 ```python
 from vision_agent.agent import VisionAgentCoderV2
-from vision_agent.agent.types import AgentMessage
+from vision_agent.models import AgentMessage
 
 agent = VisionAgentCoderV2(verbose=True)
 code_context = agent.generate_code(

docs/index.md

@@ -27,7 +27,7 @@ You can run VisionAgent in a local Jupyter Notebook [Counting cans in an image](
 You can use VisionAgent to generate code to count the number of people in an image:
 ```python
 from vision_agent.agent import VisionAgentCoderV2
-from vision_agent.agent.types import AgentMessage
+from vision_agent.models import AgentMessage
 
 agent = VisionAgentCoderV2(verbose=True)
 code_context = agent.generate_code(