TempCompass

<h2 align="center"> <a href="https://arxiv.org/abs/2403.00476">TempCompass: A benchmark to evaluate the temporal perception ability of Video LLMs</a></h2> <div align="center"> <a href='https://arxiv.org/abs/2403.00476'><img src='https://img.shields.io/badge/ArXiv-2403.00476-red'></a> <a href='https://llyx97.github.io/tempcompass/'><img src='https://img.shields.io/badge/Project-Page-Green'></a> <a href='https://huggingface.co/spaces/lyx97/TempCompass'><img src='https://img.shields.io/badge/🤗_Hugging_Face-Leaderboard-blue'></a> <a href='https://huggingface.co/datasets/lmms-lab/TempCompass'><img src='https://img.shields.io/badge/🤗_Hugging_Face-Datasets-green'></a> </div> <div> <div align="center"> <a href='https://llyx97.github.io/' target='_blank'>Yuanxin Liu^1*</a>&emsp; <a href='https://lscpku.github.io/' target='_blank'>Shicheng Li^1*</a>&emsp; <a href='https://liuyi-pku.github.io/' target='_blank'>Yi Liu¹</a>&emsp; Yuxiang Wang¹&emsp; <a href='https://renshuhuai-andy.github.io/' target='_blank'>Shuhuai Ren¹</a>&emsp; </br> <a href='https://lilei-nlp.github.io/' target='_blank'>Lei Li²</a>&emsp; <a href='https://pkucss.github.io/' target='_blank'>Sishuo Chen¹</a>&emsp; <a href='https://xusun26.github.io/' target='_blank'>Xu Sun¹</a>&emsp; <a href='https://houlu369.github.io/' target='_blank'>Lu Hou³</a> </div> <div> <div align="center"> ¹Peking University&emsp; ²The University of Hong Kong&emsp; ³Huawei Noah’s Ark Lab </div> <div align="center"> ^*Equal Contribution </div>

📢 News

[2024-10-30] 🎉🎉🎉 TempCompass is integrated into VLMEvalKit.

[2024-08-30] Results of Qwen2-VL, GPT-4o, MiniCPM-V-2.6, InternVL-2-8B, LLaVA-OneVision-Qwen-2-7B and InterLM-XComposer-2.5 are added to the leaderboard. GPT-4o establishes the new SoTA!

[2024-08-08] Results of LLaVA-Next-Video, VILA-1.5 and LongVA are added to the leaderboard.

[2024-07] 🎉🎉🎉 TempCompass is integrated into LMMs-Eval. See here for usage examples.

[2024-06-11] Result of Reka-core is added to the leaderboard.

[2024-05-25] TempCompass Leaderboard is available on HuggingFace Space 🤗.

[2024-05-16] 🎊🎊🎊 TempCompass is accepted at ACL 2024 Findings!

[2024-04-14] Evaluation result of Gemini-1.5-pro, the current SOTA Video LLM, is add.

[2024-03-23] The answer prompt is improved to better guide Video LLMs to follow the desired answer formats. The evaluation code now provides an option to disable the use of ChatGPT.

[2024-03-12] 🔥🔥🔥 The evaluation code is released now! Feel free to evaluate your own Video LLMs.

🏆 LeaderBoard

✨ Highlights

Diverse Temporal Aspects and Task Formats

• TempCompass encompasses a diverse set of temporal aspects (left) and task formats (right) to comprehensively evaluate the temporal perception capability of Video LLMs.

Conflicting Videos

• We construct conflicting videos to prevent the models from taking advantage of single-frame bias and language priors.

• 🤔 Can your Video LLM correctly answer the following question for both two videos?

  <img src="./assets/1021488277.gif" alt="Raw Video" style="float: left; width: 49%; margin-right: 10px;"> <img src="./assets/1021488277_reverse.gif" alt="Conflicting Video" style="float: left; width: 49%;">

  What is happening in the video?
  A. A person drops down the pineapple
  B. A person pushes forward the pineapple
  C. A person rotates the pineapple
  D. A person picks up the pineapple

🚀 Quick Start

To begin with, clone this repository and install some packages:

git clone https://github.com/llyx97/TempCompass.git
cd TempCompass
pip install -r requirements.txt

Data Preparation

1. Task Instructions

The task instructions can be found in questions/.

<details> <summary><span id="instruct_gen"> Task Instruction Generation Procedure </span></summary>

1. Generate Multi-Choice QA instructions (question_gen.py).

2. Manually validate quality and rectify.

3. Generate task instructions for Yes/No QA (question_gen_yes_no.py), Caption Matching (question_gen_caption_match.py) and Caption Generation (question_gen_captioning.py), based on manually rectified Multi-Choice QA instructions.

4. Manually validate quality and rectify.

</details>

2. Videos

All the processed videos can be downloaded from google drive or huggingface.

<details> <summary><span id="instruct_gen"> As an alternative, you can also download the raw videos and process them yourself </span></summary>

Run the following commands. The videos will be saved to videos/.

cd utils
python download_video.py    # Download raw videos
python process_videos.py    # Construct conflicting videos

Note: If you encounter a MoviePy error when running the processing script, please refer to this issue.

</details>

Run Inference

We use Video-LLaVA and Gemini as examples to illustrate how to conduct MLLM inference on our benchmark.

1. Video-LLaVA

Enter run_video_llava and install the environment as instructed.

Then run the following commands. The prediction results will be saved to predictions/video-llava/<task_type>.

# select <task_type> from multi-choice, yes_no, caption_matching, captioning
python inference_dataset.py --task_type <task_type>

2. Gemini

The inference script for gemini-1.5-pro is run_gemini.ipynb. It is recommended to run the script in Google Colab.

<span id="eval"> Run Evaluation </span>

After obtaining the MLLM predictions, run the following commands to conduct automatic evaluation. Remember to set your own $OPENAI_API_KEY in utils/eval_utils.py.

• Multi-Choice QA python eval_multi_choice.py --video_llm video-llava

• Yes/No QA python eval_yes_no.py --video_llm video-llava

• Caption Matching python eval_caption_matching.py --video_llm video-llava

• Caption Generation python eval_captioning.py --video_llm video-llava

Tip👉: Except for Caption Generation, you can set --disable_llm when running the scripts, which will disable chatgpt-based evaluation (i.e., entirely rely on rule-based evaluation). This is useful when you do not want to use ChatGPT API and your MLLM is good at following the instruction to generate answers of specific format.

The results of each data point will be saved to auto_eval_results/video-llava/<task_type>.json and the overall results on each temporal aspect will be printed out as follows:

{'action': 76.0, 'direction': 35.2, 'speed': 35.6, 'order': 37.7, 'attribute_change': 41.0, 'avg': 45.6}
{'fine-grained action': 58.8, 'coarse-grained action': 90.3, 'object motion': 36.2, 'camera motion': 32.6, 'absolute speed': 47.6, 'relative speed': 28.0, 'order': 37.7, 'color & light change': 43.6, 'size & shape change': 39.4, 'combined change': 41.7, 'other change': 38.9}
Match Success Rate=100.0

<span id="lmms-eval"> LMMs-Eval Evaluation </span>

Here we provide an example of how to evaluate LLaVA-Next-Video on TempCompass, using lmms-eval.

1. Clone the repo from LLaVA-Next and setup environments

git clone https://github.com/LLaVA-VL/LLaVA-NeXT
cd LLaVA-NeXT
pip install -e .

2. Run inference and evaluation in a single command

accelerate launch --num_processes 8 --main_process_port 12345 -m lmms_eval \
    --model llavavid \
    --model_args pretrained=lmms-lab/LLaVA-NeXT-Video-32B-Qwen,conv_template=qwen_1_5,video_decode_backend=decord,max_frames_num=32,mm_spatial_pool_mode=average,mm_newline_position=grid,mm_resampler_location=after \
    --tasks tempcompass \
    --batch_size 1 \
    --log_samples \
    --log_samples_suffix llava_vid_32B \
    --output_path ./logs/

You can also evaluate the performance on each task (e.g., multi-choice) seperately:

accelerate launch --num_processes 8 --main_process_port 12345 -m lmms_eval \
    --model llavavid \
    --model_args pretrained=lmms-lab/LLaVA-NeXT-Video-32B-Qwen,conv_template=qwen_1_5,video_decode_backend=decord,max_frames_num=32,mm_spatial_pool_mode=average,mm_newline_position=grid,mm_resampler_location=after \
    --tasks tempcompass_multi_choice \
    --batch_size 1 \
    --log_samples \
    --log_samples_suffix llava_vid_32B \
    --outpu