Implementation:Open compass VLMEvalKit VCRBench
Appearance
| Field | Value |
|---|---|
| source | VLMEvalKit |
| domain | Vision, Benchmarking, Video Comprehension |
Overview
Benchmark dataset implementation for VCR-Bench video comprehension and reasoning evaluation in VLMEvalKit.
Description
VCRBench inherits from VideoBaseDataset and implements the VCR-Bench benchmark for evaluating video comprehension and reasoning capabilities. The TYPE field is set to 'Video-VQA'. It provides uniformly sampled frames in chronological order with system and frame template prompts for guiding model responses.
Usage
Registered in vlmeval/dataset/__init__.py and invoked through build_dataset() by benchmark name.
Code Reference
- Source:
vlmeval/dataset/vcrbench.py, Lines: L1-302 - Import:
from vlmeval.dataset.vcrbench import VCRBench
Signature:
class VCRBench(VideoBaseDataset):
TYPE = 'Video-VQA'
SYS = 'You are an AI assistant responsible for answering questions about videos.'
...
I/O Contract
| Direction | Description |
|---|---|
| Inputs | TSV dataset file with video paths and comprehension questions |
| Outputs | Evaluation results DataFrame with VQA scores |
Usage Examples
from vlmeval.dataset import build_dataset
dataset = build_dataset('VCR-Bench')
Related Pages
Page Connections
Double-click a node to navigate. Hold to expand connections.
Principle
Implementation
Heuristic
Environment