datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
RadImageNet-VQA
RadImageNet-VQA: A Large-Scale CT and MRI Dataset for Radiologic Visual Question Answering
We introduce RadImageNet-VQA, a large-scale dataset designed for training and benchmarking radiologic VQA on CT and MRI exams. Built from the CT/MRI subset of RadImageNet and its expert-curated anatomical and pathological annotations, RadImageNet-VQA provides 750K images with 7.5M generated samples, including 750K medical captions for visual-text alignment and 6.75M… See the full description on the dataset page: https://huggingface.co/datasets/raidium/RadImageNet-VQA.SID_Set
Dataset Card for SID_Set
Dataset Summary
We provide Social media Image Detection dataSet (SID-Set), which offers three key advantages:
Extensive volume: Featuring 300K AI-generated/tampered and authentic images with comprehensive annotations.
Broad diversity: Encompassing fully synthetic and tampered images across various classes.
Elevated realism: Including images that are predominantly indistinguishable from genuine ones through mere visual inspection.
Please… See the full description on the dataset page: https://huggingface.co/datasets/RAID-techjam/SID_Set.flaird-raid-pan26RAID_none-encoded-gpt2Raid_splitbackuplienchiangprobslistMetricEval-BodyCT
MetricEval-BodyCT
This repository is a body CT benchmark for evaluating radiology report-generation metrics against
radiologists' judgment.
It covers 100 CT studies (50 chest and 50 abdomen/pelvis), with three candidate reports each. Every
candidate was independently annotated by multiple board-certified radiologists. The reference reports
are de-identified radiology reports from multiple US centers, provided by Segmed and redistributed
under the Data Use Agreement in LICENSE.… See the full description on the dataset page: https://huggingface.co/datasets/raidium/MetricEval-BodyCT.RAID-Plus
RAID+
🌐 Project Page, 🖥️ Code, 📊 Original RAID
RAID+ is an evaluation-only extension of the RAID benchmark(Dugan et al., 2024), regenerating RAID prompts using contemporary frontier models absent from the original dataset. It is intended for evaluating MGT detectors against LLMs released after RAID's publication.
This dataset was constructed as part of INSCONE: Unknown-Aware Detection of LLM-Generated Text via Informed Wild Data.
Models
Model
Samples… See the full description on the dataset page: https://huggingface.co/datasets/markstanl/RAID-Plus.clean-raid-humanatc-spotlightoculus-sft-90k-raidraid_paraphrase_testCuriaBench
🌟 Github |
📄 Paper Link |
🌐 Blog post
CuriaBench; Benchmark for the paper Curia: A Multi-Modal Foundation Model for Radiology
Usage
This repo contains evaluation datasets fro mthe Curia paper https://arxiv.org/abs/2509.06830.
You can run the curia evaluation pipeline using our github repo at https://github.com/raidium-med/curia.
The Curia-B model is available on huggingface: https://huggingface.co/raidium/curia
Acknowledgments… See the full description on the dataset page: https://huggingface.co/datasets/raidium/CuriaBench.raid-balanced-pairedraidersg-aviation-el-combined-tokenisedsequelbox_Raiden-DeepSeek-R1-Shuffled-ShareGPTimport json
from tqdm import tqdm
from datasets import load_dataset
import pandas as pd
# Example usage:
dataset = load_dataset("sequelbox/Raiden-DeepSeek-R1")["train"]
dataset = dataset.shuffle(seed=42)
output_file = "./sequelbox_Raiden-DeepSeek-R1-Shuffled-ShareGPT.parquet"
data = []
for item in tqdm(dataset):
if item["prompt"].strip() == "" or item["response"].strip() == "":
continue
data.append(
{
"conversations": [
{… See the full description on the dataset page: https://huggingface.co/datasets/PJMixers-Dev/sequelbox_Raiden-DeepSeek-R1-Shuffled-ShareGPT.sequelbox_Raiden-DeepSeek-R1-PREVIEW-Shuffled-ShareGPTlienchiangimagesam3-door-lockd1raid-modern-llmsbackupnantou1sg-aviation-el-combinedaxite-ttssam3-sam3sn_detections1sgatc-diarizeRaid_split