datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
DeepPCB-corrected
DeepPCB-corrected
PCB defect detection — cleaned version of the DeepPCB dataset.
PCB 缺陷检测 —— DeepPCB 数据集的清理修正版。
English · 中文
English
What is this?
A cleaned and consistently packaged version of the DeepPCB dataset
(Tang, Mao, Wang, Zhu & Ye — "A Robust PCB Defect Detection Method via Fusing Multiple
Hierarchical Features", arXiv:1902.06197).
DeepPCB pairs each defect-free template image with an aligned tested image carrying
bounding-box… See the full description on the dataset page: https://huggingface.co/datasets/KeenForgeAI/DeepPCB-corrected.RoboRefit-corrected
RoboRefit-corrected
RoboRefit-corrected is a locally audited derivative of the 2,000-task RoboRefIt
TestB evaluation set. It is used by eval_roborefit_mask.py for point grounding
with mask-based ground truth. It is not a replacement for, or a modification of,
the upstream RoboRefIt release.
Relationship to the upstream data
The base task order, questions, bounding boxes, and images come from the
IffYuan/Roborefit Hugging Face test split.
images/hf_XXXX.png… See the full description on the dataset page: https://huggingface.co/datasets/VLyb/RoboRefit-corrected.NEU-DET-corrected
NEU-DET-corrected
Hot-rolled steel strip surface defect detection — cleaned version of the NEU Surface Defect Database.
热轧带钢表面缺陷检测 —— NEU 表面缺陷数据库的清理修正版。
English · 中文
English
What is this?
A cleaned, complete and consistently packaged version of the
NEU Surface Defect Database (NEU-DET) from Northeastern University
(Song Kechen & Yan Yunhui).
The upstream release ships as a flat IMAGES/ + ANNOTATIONS/ (Pascal VOC) pair with
no official split and a… See the full description on the dataset page: https://huggingface.co/datasets/KeenForgeAI/NEU-DET-corrected.fr3_pickplace_extended_new_cmd_SYNC_part_correctedThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "fr3",
"total_episodes": 301,
"total_frames": 276017,
"total_tasks": 2,
"total_videos": 0,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 60,
"splits": {
"train": "0:301"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/yio-ye2004/fr3_pickplace_extended_new_cmd_SYNC_part_corrected.paper2-channels-period1-correctedfr3_pickplace_extended_new_cmd_SYNC_part_corrected_with_prompt_testThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "fr3",
"total_episodes": 301,
"total_frames": 276017,
"total_tasks": 2,
"total_videos": 0,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 60,
"splits": {
"train": "0:301"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/yio-ye2004/fr3_pickplace_extended_new_cmd_SYNC_part_corrected_with_prompt_test.TXL-PBC-corrected
TXL-PBC-corrected — Peripheral Blood Cell Detection Dataset
A corrected version of the TXL-PBC dataset (Gan, Li & Wang, Scientific Data 12, 1694, 2025).
TXL-PBC integrates four publicly available blood-cell datasets — BCCD, BCDD, PBC and
Raabin-WBC — into a single YOLO-format detection benchmark for three cell types:
WBC (white blood cell), RBC (red blood cell) and Platelets.
This release is identical to the published TXL-PBC, except that 4 duplicate images were
removed (see… See the full description on the dataset page: https://huggingface.co/datasets/KeenForgeAI/TXL-PBC-corrected.preference_data_llama_factory_corrected_format_text_onlypreference_data_llama_factory_corrected_formatraccoon-corrected
Raccoon Detection Dataset — Corrected & Verified
A fully human-reviewed, re-annotated version of the classic Raccoon object
detection dataset. Every one of the 193 images was manually reviewed; loose,
missing, and incorrect bounding boxes were corrected.
中文摘要见文末 中文版修正报告。
Dataset at a glance
Item
Value
Images
193 (.jpg)
Labels
193 (YOLO .txt)
Classes
1 — raccoon
Total boxes
211 (original: 210)
Images corrected
172 / 193
Duplicate images… See the full description on the dataset page: https://huggingface.co/datasets/KeenForgeAI/raccoon-corrected.MMSoc_Memotion_corrected
Memotion for MTEB
This repository is a byte-preserving cleanup of
mteb/MMSoc_Memotion at revision
f77e225ae55c1987b0b8cbf6badd1c10296f5f34 for
MTEB issue #5158.
The pinned source and the
official Kaggle release
both contain the same truncated train image, image_5119.png (row
4578, SHA-256 63175a3560ace1e74d4e7913206c94b6293473e09539df0e5be791df62e6a2a9). The PNG is missing part of its IDAT
payload. Permissive decoding fabricates 35 black rows, so this dataset excludes
that one… See the full description on the dataset page: https://huggingface.co/datasets/mteb/MMSoc_Memotion_corrected.GC10-DET-corrected
GC10-DET-corrected
热轧钢板表面缺陷检测数据集 · 清理修正版
A cleaned and corrected release of the GC10-DET metallic surface defect dataset
English · 中文
English
Overview
GC10-DET-corrected is a cleaned copy of the public GC10-DET dataset (surface defects on
hot-rolled steel strip, 10 classes, Pascal VOC annotations).
Image pixels and every bounding box coordinate are unchanged. Only the following were fixed:
duplicate images removed
images without annotations… See the full description on the dataset page: https://huggingface.co/datasets/KeenForgeAI/GC10-DET-corrected.comics_dataset_512_inv_manga_correctedcorrected-tobacco-dataset-with-ocr
Corrected Tobacco3482 Dataset
This repository contains a cleaned and corrected version of the Tobacco3482 dataset. It addresses significant labeling errors found in the original source that negatively impact model performance.
📌 Dataset Context
The original Tobacco3482 dataset is a staple for document image classification, but it is known to contain noisy labels. This version integrates corrections provided by the research community to ensure a higher standard for… See the full description on the dataset page: https://huggingface.co/datasets/anirudh1112/corrected-tobacco-dataset-with-ocr.PKU-Market-PCB-corrected
PKU-Market-PCB-corrected
PCB defect detection — cleaned and consistently packaged version of the PKU-Market-PCB dataset.
PCB 缺陷检测 —— PKU-Market-PCB 数据集的清理与规范化打包版。
English · 中文
English
What is this?
A cleaned and consistently packaged version of PKU-Market-PCB
(Huang, Weibo & Wei, Peng. "A PCB Dataset for Defects Detection and Classification",
arXiv:1901.08204, 2019), released by the
Open Lab on Human Robot Interaction, Peking University.
The… See the full description on the dataset page: https://huggingface.co/datasets/KeenForgeAI/PKU-Market-PCB-corrected.pick-place-three-blocks-full-emg-correctedThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "panda",
"total_episodes": 54,
"total_frames": 15183,
"total_tasks": 1,
"total_videos": 0,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 15,
"splits": {
"train": "0:54"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/jasontchan/pick-place-three-blocks-full-emg-corrected.CoCount-test-correctedpdf2image-dataset-unrevised-correctedmars-cat-episode2-corrected-assets
Mars Cat Episode 2 corrected assets
Corrected character keyframes for video generation.
log_probs_corrected_info
Corrected Results (Forced-Choice Logprobs)
Created: 2026-02-09
Why this folder exists
All Tinker/Llama eval runs before Feb 7, 2026 had a bug in how the predicted field was determined:
The greedy generation used stop=["</answer>"], which stripped the closing XML tag from the output
The XML regex <answer>(YES|NO)</answer> failed because the closing tag was missing
The fallback text extraction (_extract_yes_no_from_text) picked up "NO" from reasoning text (e.g., "NOT"… See the full description on the dataset page: https://huggingface.co/datasets/jprivera44/log_probs_corrected_info.