蚂蚁 inclusionAI 开源 ConceptEdit-12M 与 ConceptEdit-Bench
GitHub - inclusionAI/ConceptEdit: ConceptEdit-12M and ConceptEdit-Bench · GitHub
蚂蚁 inclusionAI 在 GitHub 发布 ConceptEdit-12M 与 ConceptEdit-Bench，用于大规模、基于分类体系的图像编辑数据集生成。该仓库提供三阶段流水线，并附带 ConceptBench 基准评测代码，涵盖推理、VLM 评判、指标聚合与可视化。项目采用 MIT 许可证，各步骤幂等且支持断点续跑。
正文
当前显示原文，中文全文译文尚未就绪。
inclusionAI; ConceptEdit
Repository navigation
NotificationsYou must be signed in to change notification settings; Fork0 (0); Star39 (39)You must be signed in to star a repository
ConceptEdit-12M and ConceptEdit-Bench
Repository files navigation
README; MIT license
Image-Editing Concept Pipeline
A 3-stage pipeline for generating large-scale, taxonomy-grounded image-editing datasets:
Two variants are shipped side-by-side:
Repo layout
Setup
config.py is gitignored — never commit it.
You will need:
an OpenAI-compatible VLM endpoint (e.g. vLLM or SGLang serving a vision-language model) for instruction generation and evaluation; a local FLUX checkpoint loadable by 🤗 diffusers; (optional) object-storage credentials if your source images live in object storage; local file input is fully supported as well
Running the pipeline
All commands are run from the repo root (so that config.py is on the Python path).
Single-concept
Output of each step lives next to its input:
Multi-concept
Identical commands with the multi_ prefix:
multi_instruct_gen.py can also consume a JSONL of object-storage image paths via --jsonl (one JSON object per line, with an images field). Use --help for the full list of flags.
Benchmark evaluation
The benchmark/ directory provides code for evaluating image-editing models on ConceptBench, including inference, VLM-judge evaluation, metric aggregation, and interactive visualization of evaluation reports. Please refer to benchmark/README.md for benchmark data placement, inference and evaluation commands, metric definitions, and report visualization.
Every step is idempotent and resume-safe:
instruct_gen skips images for which a JSON with the right prefix already exists; flux_edit skips JSONs whose _edit.png already exists; eval_metric skips JSONs whose _vqa_result.json already exists
Killing the process and re-running picks up exactly where it left off.
License
Released under the MIT License. See LICENSE.
ConceptEdit-12M and ConceptEdit-Bench