Silviase/CapEval

CapEvalDataset是一个用于图像描述评估和元评估的数据集。它包含多个配置，每个配置都有不同的特征和数据集大小。数据集的特征包括ID、图像、图像路径、参考描述、候选描述、评分等。数据集的准备过程涉及下载多个公开数据集（如Flickr 8k、Flickr 30k、MSCOCO val2014等），并提供了下载链接和提示。

原始信息汇总

数据集概述

1. 配置名称: composite

• 特征:
  • ID: int32
  • image: image
  • image_path: string
  • references: sequence: string
  • candidate: string
  • scores:
    • human_correctness: float32
    • human_throughness: float32
  • caption:
    • Yi-VL-6B: string
    • liuhaotian/llava-v1.5-13b: string
    • liuhaotian/llava-v1.5-7b: string
    • liuhaotian/llava-v1.6-vicuna-13b: string
    • liuhaotian/llava-v1.6-vicuna-7b: string
  • visual_context:
    • MAGAer13/mplug-owl2-llama2-7b: string
    • Yi-VL-6B: string
    • gpt4o: string
    • liuhaotian/llava-v1.5-13b: string
    • liuhaotian/llava-v1.5-7b: string
    • liuhaotian/llava-v1.6-vicuna-13b: string
    • liuhaotian/llava-v1.6-vicuna-7b: string
• 分割:
  • train:
    • num_bytes: 3137735193.686
    • num_examples: 18641
• 下载大小: 674759943
• 数据集大小: 3137735193.686

2. 配置名称: flickr8k-cf

• 特征:
  • ID: int32
  • image: image
  • image_path: string
  • references: sequence: string
  • candidate: string
  • scores:
    • human: float32
• 分割:
  • train:
    • num_bytes: 6010780139.5
    • num_examples: 47830
• 下载大小: 204593841
• 数据集大小: 6010780139.5

3. 配置名称: flickr8k-expert

• 特征:
  • ID: int32
  • image: image
  • image_path: string
  • references: sequence: string
  • candidate: string
  • scores:
    • human: float32
  • caption:
    • Yi-VL-6B: string
    • liuhaotian/llava-v1.5-13b: string
    • liuhaotian/llava-v1.5-7b: string
    • liuhaotian/llava-v1.6-vicuna-13b: string
    • liuhaotian/llava-v1.6-vicuna-7b: string
  • visual_context:
    • MAGAer13/mplug-owl2-llama2-7b: string
    • Yi-VL-6B: string
    • gpt4o: string
    • liuhaotian/llava-v1.5-13b: string
    • liuhaotian/llava-v1.5-7b: string
    • liuhaotian/llava-v1.6-vicuna-13b: string
    • liuhaotian/llava-v1.6-vicuna-7b: string
• 分割:
  • train:
    • num_bytes: 2522643541.248
    • num_examples: 16992
• 下载大小: 178628622
• 数据集大小: 2522643541.248

4. 配置名称: pascal-50s

• 特征:
  • ID: int32
  • image: image
  • image_path: string
  • references: sequence: string
  • candidate_a: string
  • candidate_b: string
  • category: int32
  • scores:
    • label: int32
  • caption:
    • Yi-VL-6B: string
    • liuhaotian/llava-v1.5-13b: string
    • liuhaotian/llava-v1.5-7b: string
    • liuhaotian/llava-v1.6-vicuna-13b: string
    • liuhaotian/llava-v1.6-vicuna-7b: string
  • visual_context:
    • MAGAer13/mplug-owl2-llama2-7b: string
    • Yi-VL-6B: string
    • gpt4o: string
    • liuhaotian/llava-v1.5-13b: string
    • liuhaotian/llava-v1.5-7b: string
    • liuhaotian/llava-v1.6-vicuna-13b: string
    • liuhaotian/llava-v1.6-vicuna-7b: string
• 分割:
  • train:
    • num_bytes: 513286199.0
    • num_examples: 4000
• 下载大小: 463428213
• 数据集大小: 513286199.0

5. 配置名称: thumb

• 特征:
  • ID: int32
  • image: image
  • image_path: string
  • references: sequence: string
  • candidate: string
  • scores:
    • human_p: float32
    • human_r: float32
    • human_total: float32
  • caption:
    • Yi-VL-6B: string
    • liuhaotian/llava-v1.5-13b: string
    • liuhaotian/llava-v1.5-7b: string
    • liuhaotian/llava-v1.6-vicuna-13b: string
    • liuhaotian/llava-v1.6-vicuna-7b: string
  • visual_context:
    • MAGAer13/mplug-owl2-llama2-7b: string
    • Yi-VL-6B: string
    • gpt4o: string
    • liuhaotian/llava-v1.5-13b: string
    • liuhaotian/llava-v1.5-7b: string
    • liuhaotian/llava-v1.6-vicuna-13b: string
    • liuhaotian/llava-v1.6-vicuna-7b: string
• 分割:
  • train:
    • num_bytes: 437942402.0
    • num_examples: 2500
• 下载大小: 86159723
• 数据集大小: 437942402.0