import os
os.system('pip install git+https://github.com/facebookresearch/detectron2.git')
os.system('mkdir checkpoints && cd checkpoints')
os.system('wget https://datarelease.blob.core.windows.net/grit/models/grit_b_densecap_objectdet.pth && cd ..')
import gradio as gr
import openai
import requests
import csv
import argparse
from models.vlog import Vlogger
parser = argparse.ArgumentParser()
parser.add_argument('--video_path', default='examples/huaqiang.mp4')
parser.add_argument('--alpha', default=10, type=int, help='Determine the maximum segment number for KTS algorithm, the larger the value, the fewer segments.')
parser.add_argument('--beta', default=1, type=int, help='The smallest time gap between successive clips, in seconds.')
parser.add_argument('--data_dir', default='./examples', type=str, help='Directory for saving videos and logs.')
parser.add_argument('--tmp_dir', default='./tmp', type=str, help='Directory for saving intermediate files.')
# * Models settings *
parser.add_argument('--openai_api_key', default='xxx', type=str, help='OpenAI API key')
parser.add_argument('--image_caption', action='store_true', dest='image_caption', default=True, help='Set this flag to True if you want to use BLIP Image Caption')
parser.add_argument('--dense_caption', action='store_true', dest='dense_caption', default=True, help='Set this flag to True if you want to use Dense Caption')
parser.add_argument('--feature_extractor', default='openai/clip-vit-base-patch32', help='Select the feature extractor model for video segmentation')
parser.add_argument('--feature_extractor_device', choices=['cuda', 'cpu'], default='cuda', help='Select the device: cuda or cpu')
parser.add_argument('--image_captioner', choices=['blip2', 'blip2-opt'], dest='captioner_base_model', default='blip2-opt', help='blip2 requires 15G GPU memory, blip requires 6G GPU memory')
parser.add_argument('--image_captioner_device', choices=['cuda', 'cpu'], default='cuda', help='Select the device: cuda or cpu, gpu memory larger than 14G is recommended')
parser.add_argument('--dense_captioner_device', choices=['cuda', 'cpu'], default='cuda', help='Select the device: cuda or cpu, < 6G GPU is not recommended>')
parser.add_argument('--audio_translator', default='large')
parser.add_argument('--audio_translator_device', choices=['cuda', 'cpu'], default='cuda')
parser.add_argument('--gpt_version', choices=['gpt-3.5-turbo'], default='gpt-3.5-turbo')
args = parser.parse_args()
def get_empty_state():
return {"total_tokens": 0, "messages": []}
def submit_api_key_fn(api_key, vlogger):
try:
vlogger.init_llm_with_api_key(api_key)
return gr.update(value = "OpenAI key submitted successful 🎉"), True, vlogger
except Exception as e:
return gr.update(value = f"Error {e}"), False, vlogger
def submit_message(prompt, state, vlogger, api_key_submitted, vlog_loaded):
if not api_key_submitted:
return gr.update(value=''), [("👀", "Please enter your OpenAI API key 😊"),], state, vlogger
if not vlog_loaded:
return gr.update(value=''), [("👀", "Please follow the instruction to select a video and generate the document for chatting 😊"),], state, vlogger
history = state['messages']
if not prompt:
return gr.update(value=''), [(history[i]['content'], history[i+1]['content']) for i in range(0, len(history)-1, 2)], state, vlogger
prompt_msg = { "role": "user", "content": prompt }
try:
history.append(prompt_msg)
answer = vlogger.chat2video(prompt)
history.append({"role": "system", "content": answer})
except Exception as e:
history.append(prompt_msg)
history.append({
"role": "system",
"content": f"Error: {e}"
})
chat_messages = [(history[i]['content'], history[i+1]['content']) for i in range(0, len(history)-1, 2)]
return '', chat_messages, state, vlogger
def clear_conversation(vlogger):
vlogger.clean_history()
# return input_message, video_inp, chatbot, vlog_outp, state, vlogger, vlog_loaded
return gr.update(value=None, visible=True), gr.update(value=None, interactive=False), None, gr.update(value=None, visible=True), get_empty_state(), vlogger, False
def vlog_fn(vid_path, vlogger, api_key_submitted):
if not api_key_submitted:
log_text = "====== Please enter your OpenAI API key first 😊 ====="
return gr.update(value=log_text, visible=True), False, vlogger
print(vid_path)
if vid_path is None:
log_text = "====== Please select an video from examples first 🤔 ====="
vloaded_flag = False
else:
log_list = vlogger.video2log(vid_path)
log_text = "\n".join(log_list)
vloaded_flag = True
return gr.update(value=log_text, visible=True), vloaded_flag, vlogger
css = """
#col-container {max-width: 90%; margin-left: auto; margin-right: auto;}
#video_inp {min-height: 300px}
#chatbox {min-height: 100px;}
#header {text-align: center;
#hint {font-size: 0.9em; padding: 0.5em; margin: 0;}
.message { font-size: 1.2em; }
"""
with gr.Blocks(css=css) as demo:
state = gr.State(get_empty_state())
vlogger = gr.State(Vlogger(args))
vlog_loaded = gr.State(False)
api_key_submitted = gr.State(False)
with gr.Column(elem_id="col-container"):
gr.Markdown("""## ChatVideo Demo
Powered by BLIP2, GRIT, Whisper, ChatGPT and LangChain
""",
elem_id="header")
gr.Markdown("*Instruction*: For the current demo, please enter OpenAI api key, select an example video, click the button to generate a document and try chatting over the video 😊", elem_id="hint")
with gr.Row():
with gr.Column(scale=6):
video_inp = gr.Video(label="video_input", interactive=False)
chatbot = gr.Chatbot(elem_id="chatbox")
input_message = gr.Textbox(show_label=False, placeholder="Enter text and press enter", visible=True).style(container=False)
btn_submit = gr.Button("Submit")
btn_clear_conversation = gr.Button("🔃 Start New Conversation")
with gr.Column(scale=6):
vlog_btn = gr.Button("Generate Video Document")
vlog_outp = gr.Textbox(label="Document output", lines=30)
with gr.Column(scale=1):
openai_api_key = gr.Textbox(
placeholder="Input OpenAI API key and press Enter",
show_label=False,
label = "OpenAI API Key",
lines=1,
type="password"
)
examples = gr.Examples(
examples=[
["examples/basketball_vlog.mp4"],
["examples/travel_in_roman.mp4"],
["examples/C8lMW0MODFs.mp4"],
["examples/outcGtbnMuQ.mp4"],
["examples/huaqiang.mp4"],
["examples/firesafety.mp4"],
],
inputs=[video_inp],
)
gr.HTML('''