# Video-LLaVA
## Overview
Video-LLaVa is an open-source multimodal LLM trained by fine-tuning LlamA/Vicuna on multimodal instruction-following data generated by Llava1.5 and VideChat. It is an auto-regressive language model, based on the transformer architecture. Video-LLaVa unifies visual representations to the language feature space, and enables an LLM to perform visual reasoning capabilities on both images and videos simultaneously.
The Video-LLaVA model was proposed in [Video-LLaVA: Learning United Visual Representation by Alignment Before Projection](https://arxiv.org/abs/2311.10122) by Bin Lin, Yang Ye, Bin Zhu, Jiaxi Cui, Munang Ning, Peng Jin, Li Yuan.
The abstract from the paper is the following:
*The Large Vision-Language Model (LVLM) has enhanced the performance of various downstream tasks in
visual-language understanding. Most existing approaches
encode images and videos into separate feature spaces,
which are then fed as inputs to large language models.
However, due to the lack of unified tokenization for images and videos, namely misalignment before projection, it
becomes challenging for a Large Language Model (LLM)
to learn multi-modal interactions from several poor projection layers. In this work, we unify visual representation into the language feature space to advance the foundational LLM towards a unified LVLM. As a result, we establish a simple but robust LVLM baseline, Video-LLaVA,
which learns from a mixed dataset of images and videos,
mutually enhancing each other. Video-LLaVA achieves superior performances on a broad range of 9 image benchmarks across 5 image question-answering datasets and 4
image benchmark toolkits. Additionally, our Video-LLaVA
also outperforms Video-ChatGPT by 5.8%, 9.9%, 18.6%,
and 10.1% on MSRVTT, MSVD, TGIF, and ActivityNet, respectively. Notably, extensive experiments demonstrate that
Video-LLaVA mutually benefits images and videos within
a unified visual representation, outperforming models designed specifically for images or videos. We aim for this
work to provide modest insights into the multi-modal inputs
for the LLM*
## Usage tips:
- We advise users to use padding_side="left" when computing batched generation as it leads to more accurate results. Simply make sure to call processor.tokenizer.padding_side = "left" before generating.
- Note the model has not been explicitly trained to process multiple images/videos in the same prompt, although this is technically possible, you may experience inaccurate results.
- Note that the video inputs should have exactly 8 frames at the input, since the models were trained in that setting.
This model was contributed by [RaushanTurganbay](https://huggingface.co/RaushanTurganbay).
The original code can be found [here](https://github.com/PKU-YuanGroup/Video-LLaVA).
> [!NOTE]
> LLaVA models after release v4.46 will raise warnings about adding `processor.patch_size = {{patch_size}}`, `processor.num_additional_image_tokens = {{num_additional_image_tokens}}` and processor.vision_feature_select_strategy = {{vision_feature_select_strategy}}`. It is strongly recommended to add the attributes to the processor if you own the model checkpoint, or open a PR if it is not owned by you.
Adding these attributes means that LLaVA will try to infer the number of image tokens required per image and expand the text with as many `` placeholders as there will be tokens. Usually it is around 500 tokens per image, so make sure that the text is not truncated as otherwise there will be failure when merging the embeddings.
The attributes can be obtained from model config, as `model.config.vision_config.patch_size` or `model.config.vision_feature_select_strategy`. The `num_additional_image_tokens` should be `1` if the vision backbone adds a CLS token or `0` if nothing extra is added to the vision patches.
## Usage example
### Single Media Mode
The model can accept both images and videos as input. Here's an example code for inference in half-precision (`torch.float16`):
```python
import av
import torch
import numpy as np
from transformers import VideoLlavaForConditionalGeneration, VideoLlavaProcessor
def read_video_pyav(container, indices):
'''
Decode the video with PyAV decoder.
Args:
container (`av.container.input.InputContainer`): PyAV container.
indices (`List[int]`): List of frame indices to decode.
Returns:
result (np.ndarray): np array of decoded frames of shape (num_frames, height, width, 3).
'''
frames = []
container.seek(0)
start_index = indices[0]
end_index = indices[-1]
for i, frame in enumerate(container.decode(video=0)):
if i > end_index:
break
if i >= start_index and i in indices:
frames.append(frame)
return np.stack([x.to_ndarray(format="rgb24") for x in frames])
# Load the model in half-precision
model = VideoLlavaForConditionalGeneration.from_pretrained("LanguageBind/Video-LLaVA-7B-hf", torch_dtype=torch.float16, device_map="auto")
processor = VideoLlavaProcessor.from_pretrained("LanguageBind/Video-LLaVA-7B-hf")
# Load the video as an np.arrau, sampling uniformly 8 frames
video_path = hf_hub_download(repo_id="raushan-testing-hf/videos-test", filename="sample_demo_1.mp4", repo_type="dataset")
container = av.open(video_path)
total_frames = container.streams.video[0].frames
indices = np.arange(0, total_frames, total_frames / 8).astype(int)
video = read_video_pyav(container, indices)
# For better results, we recommend to prompt the model in the following format
prompt = "USER: 
