Implementation:Datajuicer Data juicer VideoUndistortMapper

Knowledge Sources	Datajuicer_Data_juicer
Domains	Data_Processing, Mapping
Last Updated	2026-02-14 16:00 GMT

Overview

Concrete tool for removing lens distortion from raw videos provided by Data-Juicer.

Description

VideoUndistortMapper reads per-frame camera intrinsics and distortion coefficients from sample metadata, computes undistortion remap matrices using OpenCV's omnidir module (CMei's model), applies the remapping to each frame in batches, saves intermediate results as temporary TS files, and concatenates them into a final MP4 using FFmpeg with configurable CRF quality. The operator requires the sample to include 'intrinsics' and 'xi' fields and optionally accepts 'distortion_coefficients', 'rotation_matrix', and 'intrinsics_new'.

Usage

Use when you need to correct optical lens distortion artifacts from raw video data, producing geometrically accurate video needed for 3D reconstruction, SLAM, and other geometric computer vision tasks.

Code Reference

Source Location

• Repository: Datajuicer_Data_juicer
• File: data_juicer/ops/mapper/video_undistort_mapper.py

Signature

@OPERATORS.register_module("video_undistort_mapper")
class VideoUndistortMapper(Mapper):
    def __init__(self, output_video_dir: str = DATA_JUICER_ASSETS_CACHE,
                 tag_field_name: str = MetaKeys.video_undistortion_tags,
                 batch_size_each_video: int = 1000,
                 crf: int = 22,
                 *args, **kwargs):

Import

from data_juicer.ops.mapper.video_undistort_mapper import VideoUndistortMapper

I/O Contract

Inputs

Name	Type	Required	Description
output_video_dir	str	No	Output directory to save undistorted videos. Default: DATA_JUICER_ASSETS_CACHE
tag_field_name	str	No	Field name to store the undistortion tags. Default: "video_undistortion_tags"
batch_size_each_video	int	No	Number of frames to process per temporary TS file batch. Default: 1000
crf	int	No	Constant Rate Factor for FFmpeg encoding quality. Default: 22

Outputs

Name	Type	Description
sample[Fields.meta][tag_field_name]	dict	Contains 'new_video_path' pointing to the undistorted MP4 output file

Usage Examples

process:
  - video_undistort_mapper:
      output_video_dir: "./undistorted_videos"
      crf: 18
      batch_size_each_video: 500

Related Pages

• Environment:Datajuicer_Data_juicer_Python_Runtime_Environment