# this is .py for store constants 
MODEL_INFO = ["Model", "Model Type", "Model Size", "Frames"]

TASK_INFO = ["Avg. All", "Avg. Action", "Avg. Direction", "Avg. Speed", "Avg. Event Order", "Avg. Attribute Change", "Avg. Multi-Choice", "Avg. Yes/No", "Avg. Caption Matching", "Avg. Caption Generation",
                "Action. Multi-Choice", "Action. Yes/No", "Action. Caption Matching", "Action. Caption Generation", 
                "Direction. Multi-Choice", "Direction. Yes/No", "Direction. Caption Matching", "Direction. Caption Generation",
                "Speed. Multi-Choice", "Speed. Yes/No", "Speed. Caption Matching", "Speed. Caption Generation",
                "Event Order. Multi-Choice", "Event Order. Yes/No", "Event Order. Caption Matching", "Event Order. Caption Generation",
                "Attribute Change. Multi-Choice", "Attribute Change. Yes/No", "Attribute Change. Caption Matching", "Attribute Change. Caption Generation",
                "Notes"]

AVG_INFO = ["Avg. All", "Avg. Action", "Avg. Direction", "Avg. Speed", "Avg. Event Order", "Avg. Attribute Change", "Avg. Multi-Choice", "Avg. Yes/No", "Avg. Caption Matching", "Avg. Caption Generation"]
DATA_TITILE_TYPE = ["markdown", "markdown", "markdown", "markdown",
                    "number", "number", "number", "number", "number", "number", "number", "number", "number", "number",
                    "number", "number", "number", "number",
                    "number", "number", "number", "number",
                    "number", "number", "number", "number",
                    "number", "number", "number", "number",
                    "number", "number", "number", "number",
                    "markdown"]
CSV_DIR = "./file/result.csv"

# COLUMN_NAMES = MODEL_INFO + TASK_INFO
COLUMN_NAMES = MODEL_INFO + TASK_INFO

LEADERBORAD_INTRODUCTION = """
Welcome to the leaderboard of TempCompass! 🏆

TempCompass is a benchmark to evaluate the temporal perception ability of Video LLMs. It consists of 410 videos and 7,540 task instructions, covering 11 temporal aspects and 4 task types. Please refer to [our paper](https://arxiv.org/abs/2403.00476) for more details.
"""

SUBMIT_INTRODUCTION = """
# TempCompass Leaderboard

Welcome to the leaderboard of the TempCompass! 🏆

## Submit Instruction

1.Run inference and automatic evaluation according to our [github repository](https://github.com/llyx97/TempCompass?tab=readme-ov-file#-quick-start). You will obtain the JSON file `<task_type>.json`, where `<task_type>` correspond to one of the four categories: `multi-choice`, `yes_no`, `caption_matching` and `captioning`. (Example files can be found [here](https://huggingface.co/spaces/lyx97/TempCompass/tree/main/file/example_eval_results))

2. Merge the four JSON files using [merge_eval_result.py](https://huggingface.co/spaces/lyx97/TempCompass/blob/main/merge_eval_result.py) and obtain `merged_result.json` (an example can be found [here](https://huggingface.co/spaces/lyx97/TempCompass/blob/main/file/example_eval_results/merged_result.json)).


### Submit Example
For example, if you want to submit Video-LLaVA-7B's result in the leaderboard, you need to:
1. Fill in ‘Video-LLaVA-7B’ in ‘Model Name’ if it is your first time to submit your result (You can leave ‘Revision Model Name’ blank).
2. Fill in ‘Video-LLaVA-7B’ in ‘Revision Model Name’ if you want to update your result (You can leave ‘Model Name’ blank).
3. Select ‘VideoLLM’ in ‘Model Type’.
4. Fill in ‘https://huggingface.co/LanguageBind/Video-LLaVA-7B’ in ‘Model Link’.
5. Fill in ‘7B’ in ‘Model size’.
6. Upload `merged_result.json`.
7. Click the ‘Submit Eval’ button.
8. Click ‘Refresh’ to obtain the uploaded leaderboard.

## If you have any questions, please contact [liuyuanxin@stu.pku.edu.cn](liuyuanxin@stu.pku.edu.cn).
"""

# """
# For `multi-choice`, `yes_no`, `caption_matching`, the evaluation result of each question contains five keys. A specific example is as follows:
# ```python
# {
#     "question": "What activity is the monkey engaged in?\\nA. swimming\\nB. running\\nC. climbing\\nD. fighting",
#     "gt-answer": "D. fighting",
#     "video-llm-prediction": "D",
#     "match_success": true,  # whether the video-llm-prediction can be assessed by rule-based matching
#     "rating": 1
# }
# ```

# For `captioning`, we prompt chatgpt to answer the multi-choice question, using the Video LLM generated caption as context. An example of evalution result is as follows:
# ```python
# {
#     "chatgpt-reasoning": "The video description specifically mentions that the man is dribbling a basketball, dunking a basketball, and passing a basketball.",
#     "chatgpt-answer": "B. dribbling a basketball, C. passing a basketball",
#     "video-llm-prediction": "The video showcases a man dribbling a basketball, dunking a basketball, and passing a basketball. The man is seen moving around the court while performing these actions. The video captures the man's movements and the sound of the ball bouncing on the court. The man's dribbling skills are impressive, and he seems to be in control of the ball at all times. The dunking and passing actions are also executed with precision, and the man's movements are fluid and graceful. Overall, the video is a great display of basketball skills and is sure to impress any basketball",
#     "gt-answer": "A. dunking a basketball",
#     "rating": 0
# }
# ```
# """

TABLE_INTRODUCTION = """In the table below, we summarize each task performance of all the models.
        We use accurancy(%) as the primary evaluation metric for each tasks. 
    """

LEADERBORAD_INFO = """
      Based on powerful Large Language Models (LLMs), recent generative Multimodal Large Language Models (MLLMs) have gained prominence as a pivotal research area, exhibiting remarkable capability for both comprehension and generation. 
      In this work, we address the evaluation of generative comprehension in MLLMs as a preliminary step towards a comprehensive assessment of generative models, by introducing a benchmark named SEED-Bench. 
      SEED-Bench consists of 19K multiple choice questions with accurate human annotations (x6 larger than existing benchmarks), which spans 12 evaluation dimensions including the comprehension of both the image and video modality. 
      We develop an advanced pipeline for generating multiple-choice questions that target specific evaluation dimensions, integrating both automatic filtering and manual verification processes. 
      Multiple-choice questions with groundtruth options derived from human annotation enables an objective and efficient assessment of model performance, eliminating the need for human or GPT intervention during evaluation. 
      We further evaluate the performance of 18 models across all 12 dimensions, covering both the spatial and temporal understanding. 
      By revealing the limitations of existing MLLMs through evaluation results, we aim for SEED-Bench to provide insights for motivating future research.
"""

CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
CITATION_BUTTON_TEXT = r"""
@article{liu2024tempcompass,
  title   = {TempCompass: Do Video LLMs Really Understand Videos?},
  author  = {Yuanxin Liu and Shicheng Li and Yi Liu and Yuxiang Wang and Shuhuai Ren and Lei Li and Sishuo Chen and Xu Sun and Lu Hou},
  year    = {2024},
  journal = {arXiv preprint arXiv: 2403.00476}
}
"""