AnalysisAiBot

[Quick start] [Setting] pip install -r requirements.txt <-install packages AnalysisAibot.py <- this is most recent version main.py <- fast api

Default directory structure for application (the paths can be changed during the app execution process):

Need to generate api_secrets.toml file put api key, and image_sample folder in documents, if you use default directory

your_folder/
├── main.py
├── api_secrets.toml
├── templates/
│   └── index.html
├── documents/
│   └── image_sample
│       └── Images1
│           ├── video_F001.png
│           └── .... (other image files)
│   └── json/
│       ├── visual_tag_en.json
│       ├── visual_tag_kor.json
│       ├── response_sample_en.json
│       ├── response_sample_kor.json
└── results/ 

how to excute

[Sstreamlit version] streamlit run AnalysisAibot.py on terminal (need to install streamlit before running this)

[FastApi version] uvicorn main:app --reload --port 8000 (8000 is port number)

                ->  http://127.0.0.1:8000/docs     (8000 is port number)

1. MultipleimgsAi (Prompt reinforcement learning for better performance )

oldversion - imgAi.py (single img)

MultipleImgsAi_V0.1.py -> Multiple img

MultipleImgsAi_V0.2.py -> write results on json,txt files , and set reasonable max_token value by variable

MultipleImgsAi_V0.3.py -> Korean version tag, increase maxtokens per each img 800 -> 1100

MultipleImgsAi_V0.4.py -> enhance prompt (optimize + control simple case (ex: only logo) )

MultipleImgsAi_V0.5.py -> add Cost Cacluation (token calculator), fix encoding (but answers become simple)

MultipleImgsAi_V0.6.py (second goal done) -> imgs->textExplanation(until 0.5v) , imgs->textExplanation->Tag(0.6v~)!!

MultipleImgsAi_v0.7.py -> 1. normal python -> streamlit python 2. option eng,kor version (template % answer) 3. control bar (temperature (default =0.3)) (After v0.7, AnaalysisBot and MultipleImgsAi are integreted)

todolist -> 1. reduce hallucination (Find optimal temperature & enhance prompt)

2. AnalysisAibot

Anaylsis images from short videos

demoaivison.py -> convert all image files to base64 and review

AnalysisAiBot_V0.1.py -> add json template directly (but still get different format answer)

AnalysisAiBot_V0.2.py -> answer better form, but still not follow format

AnalysisAiBot_V0.3.py -> bring json contents from json files -> use one-shot tech to improve the answer formatting (but the given answer format also made form gpt not human so need to be changed) (and one-shot is not enough, two or three shot..) Increase max token 3000 -> 4000 (input img max is 30~39 [predict:35])

AnalysisAiBot_V0.4.py (AnaylsisAiBotStreamlit.py) -> same results,but now can change the images folder, change template, saved json file name streamlit run name.py

AnalysisAiBot_V0.5.py (AnaylsisAiBotStreamlit.py is not 0.5v, but experiment version):

Fix the critical bugs that bring the wrong order files from path. -> previous (lesser than 20 percent accurancy) - > near to 60 percent.

AnalysisAiBot_V0.6.py

until 0.5V , img -> text(explanation) -> tag 0.6v img-> tag (direct)

AnalysisAib0t_V0.7.py = MultipleImgsAi_v0.7.py

AnalysisAib0t_V0.8.py -> defalut temperature 0.3->0.25, make integra function to make it readable.

3. Main.py (FastAPi)

V0.1 -> just conver to Fast api (no main page) V0.2 -> main page (mini test) V0.3 -> temperature default changed, use integrated function

4. others

demoaivision.py -> basic api how to use img file

imagepath.py -> path printer (see order of the files)

imgAi.py -> single img with simple prompt