alibaba / alimama-video-narrator Goto Github PK

Video storytelling is engaging multimedia content that utilizes video and its accompanying narration to share a story and attract the audience, where a key challenge is creating narrations for recorded visual scenes. Previous studies on dense video captioning and video story generation have made some progress. However, in practical applications, we typically require synchronized narrations for ongoing visual scenes.

In this work, we introduce a new task of Synchronized Video Storytelling, which aims to generate synchronous and informative narrations for videos. These narrations, associated with each video clip, should relate to the visual content, integrate relevant knowledge, and have an appropriate word count corresponding to the clip's duration. Specifically, a structured storyline is beneficial to guide the generation process, ensuring coherence and integrity. To support the exploration of this task, we introduce a new benchmark dataset E-SyncVidStory with rich annotations. Since existing Multimodal LLMs are not effective in addressing this task in one-shot or few-shot settings, we propose a framework named VideoNarrator that can generate a storyline for input videos and simultaneously generate narrations with the guidance of the generated or predefined storyline. We further introduce a set of evaluation metrics to thoroughly assess the generation. Both automatic and human evaluations validate the effectiveness of our approach.

• 2024/12: Video frame features and models can be accessed via Onedrive.
• 2024/07: Our code and dataset annotations are released. Video features will be available soon.

• Abstract
• Release 📢
• Contents
• Install
• Dataset
• Train
• Inference
• Evaluation
• Citation

1. Clone this repository

git clone https://github.com/alibaba/alimama-video-narrator
cd alimama-video-narrator

pip install --upgrade pip
conda env create -f environment.yml

cd data_process/
python process_video.py
python get_blip_fea.py

cd data_process/
# Visual Compression & Memory Consolidation
python get_training_data.py ./blip_fea/video_cuts/ ../data/all_video_data.json 
cp training_data.json ../data/split/

cd ../data/split/
python split.py training_data.json ../all_video_data.json
python get_cut_data.py train.json train_shots.json

We apply the pretraining model firefly-baichuan-7b, with the details shown in: https://github.com/yangjianxin1/Firefly
You can directly use the baichuan-7b model, downloaded from: https://huggingface.co/baichuan-inc/Baichuan-7B

bash train.sh

bash infer.sh

python tokenize_output.py $chk_path/output.json
cd metrics/evaluator_for_caption/
python evaluate_ads.py $chk_path/out_tokens.json

Download Chinese_CLIP model from: https://huggingface.co/OFA-Sys/chinese-clip-vit-base-patch16

cd metrics/EMScore/
python eval_ad_with_ref.py --inpath $chk_path/output.json

Download chinese-roberta-large model from: https://huggingface.co/hfl/chinese-roberta-wwm-ext-large

cd metrics/roberta_based/
# info_sim
python info_sim.py ../data/all_video_data.json $chk_path/output.json idf_with_all_ref.json
# info_diverse
python info_diverse.py $chk_path/output.json idf_with_all_ref.json 

cd metrics/roberta_based/
python count_intra_repeat.py chk_path/output.json

@misc{yang2024synchronizedvideostorytellinggenerating,
      title={Synchronized Video Storytelling: Generating Video Narrations with Structured Storyline}, 
      author={Dingyi Yang and Chunru Zhan and Ziheng Wang and Biao Wang and Tiezheng Ge and Bo Zheng and Qin Jin},
      year={2024},
      eprint={2405.14040},
      archivePrefix={arXiv},
      primaryClass={cs.MM},
      url={https://arxiv.org/abs/2405.14040}, 
}