JaronTHU/Video-CCAM-9B-v1.1

Model Summary

Video-CCAM-9B-v1.1 is a lightweight Video-MLLM developed by TencentQQ Multimedia Research Team.

Usage

Inference using Huggingface transformers on NVIDIA GPUs. Requirements tested on python 3.9/3.10.

pip install -U pip torch transformers peft decord pysubs2 imageio

Inference

import os
import torch
from PIL import Image
from transformers import AutoModel

from eval import load_decord

os.environ['TOKENIZERS_PARALLELISM'] = 'false'

videoccam = AutoModel.from_pretrained(
    '<your_local_path_1>',
    trust_remote_code=True,
    torch_dtype=torch.bfloat16,
    device_map='auto',
    _attn_implementation='flash_attention_2',
    # llm_name_or_path='<your_local_llm_path>',
    # vision_encoder_name_or_path='<your_local_vision_encoder_path>'
)


messages = [
    [
        {
            'role': 'user',
            'content': '<image>\nDescribe this image in detail.'
        }
    ], [
        {
            'role': 'user',
            'content': '<video>\nDescribe this video in detail.'
        }
    ]
]

images = [
    Image.open('assets/example_image.jpg').convert('RGB'),
    load_decord('assets/example_video.mp4', sample_type='uniform', num_frames=32)
]

response = videoccam.chat(messages, images, max_new_tokens=512, do_sample=False)

print(response)

Please refer to Video-CCAM for more details.

Benchmarks

Benchmark	Video-CCAM-9B	Video-CCAM-9B-v1.1
MVBench (32 frames)	61.08	64.60
MSVD-QA (32 frames)	76.9/4.1	77.9/4.2
MSRVTT-QA (32 frames)	58.7/3.5	65.9/3.8
ActivityNet-QA (32 frames)	56.2/3.6	58.7/3.8
TGIF-QA (32 frames)	83.9/4.4	84.0/4.5
Video-MME (w/o sub, 96 frames)	49.4	50.3
Video-MME (w sub, 96 frames)	55.2	52.6
MLVU (M-Avg, 96 frames)	59.4	58.5
MLVU (G-Avg, 96 frames)	3.91	3.98
VideoVista (96 frames)	64.39	69.00

• The accuracies and scores of MSVD-QA,MSRVTT-QA,ActivityNet-QA,TGIF-QA are evaluated by gpt-3.5-turbo-0125.

Acknowledgement

• xtuner: Video-CCAM-9B is trained using the xtuner framework. Thanks for their excellent works!
• Yi-1.5-9B-Chat: Great language models developed by 01.AI.
• SigLIP SO400M: Outstanding vision encoder developed by Google.

License

The model is licensed under the MIT license.