🎞 VLog: Video as a Long Document

Given a long video, we turn it into a doc containing visual + audio info. By sending this doc to ChatGPT, we can chat over the video!

News

• 23/April/2023: We release Huggingface gradio demo!
• 20/April/2023: We release our project on github and local gradio demo!

To Do List

Done

• [x] LLM Reasoner: ChatGPT (multilingual) + LangChain
• [x] Vision Captioner: BLIP2 + GRIT
• [x] ASR Translator: Whisper (multilingual)
• [x] Video Segmenter: KTS
• [x] Huggingface Space

Doing

• [ ] Optimize the codebase efficiency
• [ ] Improve Vision Models: MiniGPT-4 / LLaVA, Family of Segment-anything
• [ ] Improve ASR Translator for better alignment
• [ ] Introduce Temporal dependency
• [ ] Replace ChatGPT with own trained LLM

🧸 Examples

[ News - GPT4 launch event ]

[ TV series - 征服之华强买瓜 ]

[ TV series - The Big Bang Theory ]

[ Travel video - Travel in Rome ]

[ Vlog - Basketball training ]

🔨 Preparation

Please find installation instructions in install.md.

🌟 Start here

Run in cmd

python main.py --video_path examples/buy_watermelon.mp4 --openai_api_key xxxxx

The generated video document will be generated and saved in examples/buy_watermelon.log

Run in Gradio

python main_gradio.py --openai_api_key xxxxx

🙋 Suggestion

Stay tuned for our project 🔥

If you have more suggestions or functions need to be implemented in this codebase, feel free to drop us an email [email protected], [email protected] or open an issue.

😊 Acknowledgment

This work is based on ChatGPT, BLIP2, GRIT, KTS, Whisper, LangChain, Image2Paragraph.