JensParslov/BenchPress

BenchPress是SynthEdit的公共基准测试版本，仅包含面向基准测试的资产：一个便携式parquet文件以及每行引用的配对编辑前后图像。数据集包含1500个基准测试行，2558张图像，总计6.40 GiB的图像数据。源parquet文件路径为`/zhome/ca/9/146686/synthedit2/data/eval_mc_negatives.parquet`。布局包括`benchpress.parquet`（便携式元数据parquet文件，包含数据集相对图像路径）和`images/<sample_id>/before.png`、`images/<sample_id>/after.png`（图像文件）。使用方法包括下载数据集快照，并将基准测试运行器指向本地快照根目录。数据集还附带了一个Croissant JSON-LD清单`croissant.json`，用于机器可读的数据集发现和验证。

BenchPress is the public benchmark release for SynthEdit. It contains the benchmark-facing assets only: a portable parquet plus the paired before/after images referenced by each row. The dataset includes 1500 benchmark rows, 2558 images, and 6.40 GiB of image data. The source parquet is located at `/zhome/ca/9/146686/synthedit2/data/eval_mc_negatives.parquet`. The layout consists of `benchpress.parquet` (portable metadata parquet with dataset-relative image paths) and `images/<sample_id>/before.png`, `images/<sample_id>/after.png` (image files). Usage involves downloading the dataset snapshot and pointing your benchmark runner at the local snapshot root. The dataset is also published with a Croissant JSON-LD manifest at `croissant.json` for machine-readable dataset discovery and validation.