Implementation:Open compass VLMEvalKit EgoExoBench Utils
| Field | Value |
|---|---|
| source | VLMEvalKit |
| domain | Vision, Benchmarking, Egocentric-Exocentric Video Utilities |
Overview
Utility module providing evaluation metrics, answer extraction, and video data augmentation transforms for the EgoExoBench benchmark in VLMEvalKit.
Description
This utility file does not define a dataset class. It provides helper functions including get_dimension_rating for computing category-level accuracy, extract_characters_regex for parsing multiple-choice answers from model responses, and a comprehensive set of video augmentation transform classes (GroupRandomCrop, GroupCenterCrop, GroupScale, GroupNormalize, etc.) used for video frame preprocessing.
Usage
Imported by vlmeval/dataset/EgoExoBench/egoexobench.py via wildcard import to support the EgoExoBench dataset evaluation.
Code Reference
- Source:
vlmeval/dataset/EgoExoBench/utils.py, Lines: L1-771 - Import:
from vlmeval.dataset.EgoExoBench.utils import *
Key Functions and Classes:
def get_dimension_rating(data_path, category_type='subtask_type'):
...
def extract_characters_regex(s):
...
class GroupRandomCrop(object):
...
class GroupCenterCrop(object):
...
class GroupNormalize(object):
...
I/O Contract
| Direction | Description |
|---|---|
| Inputs | Data paths to evaluation results, model response strings, video frames |
| Outputs | Accuracy dictionaries, extracted answer characters, transformed video frames |
Related Pages
Page Connections
Double-click a node to navigate. Hold to expand connections.
Principle
Implementation
Heuristic
Environment