mlpc-lab/BLIVA
9
1import argparse2import os3import random4 5import numpy as np6import torch7import torch.backends.cudnn as cudnn8import gradio as gr9 10from bliva.common.config import Config11from bliva.common.dist_utils import get_rank12from bliva.common.registry import registry13from bliva.conversation.conversation import Chat, CONV_VISION, CONV_DIRECT14 15# imports modules for registration16 17from bliva.models import *18from bliva.processors import *19from bliva.models import load_model_and_preprocess20from evaluate import disable_torch_init21 22def parse_args():23 parser = argparse.ArgumentParser(description="Demo")24 parser.add_argument("--model_name",default='bliva_vicuna', type=str, help='model name')25 parser.add_argument("--gpu_id", type=int, default=0, help="specify the gpu to load the model.")26 args = parser.parse_args()27 return args28 29# ========================================30# Model Initialization31# ========================================32 33print('Initializing Chat')34args = parse_args()35 36if torch.cuda.is_available():37 device='cuda:{}'.format(args.gpu_id)38else:39 device=torch.device('cpu')40 41disable_torch_init()42if args.model_name == "blip2_vicuna_instruct":43 model, vis_processors, _ = load_model_and_preprocess(name=args.model_name, model_type="vicuna7b", is_eval=True, device=device)44elif args.model_name == "bliva_vicuna":45 model, vis_processors, _ = load_model_and_preprocess(name=args.model_name, model_type="vicuna7b", is_eval=True, device=device)46elif args.model_name == "bliva_flant5":47 model, vis_processors, _ = load_model_and_preprocess(name=args.model_name, model_type="flant5xxl", is_eval=True, device=device)48else:49 print("Model not found") 50 51vis_processor = vis_processors["eval"]52 53 54# vis_processor_cfg = cfg.datasets_cfg.cc_sbu_align.vis_processor.train55# vis_processor = registry.get_processor_class(vis_processor_cfg.name).from_config(vis_processor_cfg)56chat = Chat(model, vis_processor, device=device)57print('Initialization Finished')58 59# ========================================60# Gradio Setting61# ========================================62 63def gradio_reset(chat_state, img_list):64 if chat_state is not None:65 chat_state.messages = []66 if img_list is not None:67 img_list = []68 return None, gr.update(value=None, interactive=True), gr.update(placeholder='Please upload your image first', interactive=False),gr.update(value="Upload & Start Chat", interactive=True), chat_state, img_list69 70def upload_img(gr_img, text_input, chat_state):71 if gr_img is None:72 return None, None, gr.update(interactive=True), chat_state, None73 chat_state = CONV_DIRECT.copy() #CONV_VISION.copy()74 img_list = []75 llm_message = chat.upload_img(gr_img, chat_state, img_list)76 return gr.update(interactive=False), gr.update(interactive=True, placeholder='Type and press Enter'), gr.update(value="Start Chatting", interactive=False), chat_state, img_list77 78def gradio_ask(user_message, chatbot, chat_state):79 if len(user_message) == 0:80 return gr.update(interactive=True, placeholder='Input should not be empty!'), chatbot, chat_state81 chat.ask(user_message, chat_state)82 chatbot = chatbot + [[user_message, None]]83 return '', chatbot, chat_state84 85 86def gradio_answer(chatbot, chat_state, img_list, num_beams, temperature):87 llm_message = chat.answer(conv=chat_state,88 img_list=img_list,89 num_beams=num_beams,90 temperature=temperature,91 max_new_tokens=300,92 max_length=2000)[0]93 chatbot[-1][1] = llm_message[0]94 return chatbot, chat_state, img_list95 96title = """<h1 align="center">Demo of BLIVA</h1>"""97description = """<h3>This is the demo of BLIVA. Upload your images and start chatting!. <br> To use 98 example questions, click example image, hit upload, and press enter in the chatbox.</h3>"""99article = """<p><a href='https://gordonhu608.github.io/bliva/'><img src='https://img.shields.io/badge/Project-Page-Green'></a></p><p><a href='https://github.com/mlpc-ucsd/BLIVA'><img src='https://img.shields.io/badge/Github-Code-blue'></a></p><p><a href='https://arxiv.org/abs/2308.09936'><img src='https://img.shields.io/badge/Paper-ArXiv-red'></a></p>100"""101 102#TODO show examples below103 104with gr.Blocks() as demo:105 gr.Markdown(title)106 gr.Markdown(description)107 gr.Markdown(article)108 109 with gr.Row():110 with gr.Column(scale=0.5):111 image = gr.Image(type="pil")112 upload_button = gr.Button(value="Upload & Start Chat", interactive=True, variant="primary")113 clear = gr.Button("Restart 🔄")114 115 num_beams = gr.Slider(116 minimum=1,117 maximum=10,118 value=5,119 step=1,120 interactive=True,121 label="beam search numbers)",122 )123 124 temperature = gr.Slider(125 minimum=0.1,126 maximum=2.0,127 value=1.0,128 step=0.1,129 interactive=True,130 label="Temperature",131 )132 133 with gr.Column():134 chat_state = gr.State()135 img_list = gr.State()136 chatbot = gr.Chatbot(label='BLIVA')137 text_input = gr.Textbox(label='User', placeholder='Please upload your image first', interactive=False)138 139 gr.Examples(examples=[140 [f"images/example.jpg", "Describe this image in detail."],141 [f"images/img3.jpg", "What is this image about?"],142 [f"images/img4.jpg", "What is the title of this movie?"],143 ], inputs=[image, text_input]) 144 145 upload_button.click(upload_img, [image, text_input, chat_state], [image, text_input, upload_button, chat_state, img_list])146 147 text_input.submit(gradio_ask, [text_input, chatbot, chat_state], [text_input, chatbot, chat_state]).then(148 gradio_answer, [chatbot, chat_state, img_list, num_beams, temperature], [chatbot, chat_state, img_list]149 )150 clear.click(gradio_reset, [chat_state, img_list], [chatbot, image, text_input, upload_button, chat_state, img_list], queue=False)151 152demo.launch(enable_queue=True)