dataset

Bilingual

How to easily dig through the Epstein files yourself
2 months ago
- Prem Thakker and the author published an article in Zeteo analyzing the latest Epstein dataset released by the House Oversight Committee.
- The dataset is 26,000 documents, 40GB in size, implicating elites like Trump, Bannon, Thiel, and Summers.
- Zeteo created a searchable version using Google Pinpoint, which indexes documents and allows entity extraction by names, organizations, locations, and timestamps.
- Users can search the dataset themselves, e.g., by phrases like 'dog that hasn't barked' to find Epstein's email about Trump.
- Multiple filters can be combined, such as filtering by both 'Jeffrey Epstein' and 'Steven Bannon' to see 79 documents mentioning both.
- The author encourages readers to contact them on Signal if they find anything interesting.
Automated sign detection across the Electronic Babylonian Library
2 months ago
- The paper introduces the largest annotated cuneiform sign dataset and uses a Deformable Detection Transformer (DETR) for object detection.
- The model is evaluated on 173 and 106 class granularities, achieving 28-37% improvement over prior work on COCO-style metrics.
- The system integrates automatic tablet-side extraction, heuristic line grouping, and n-gram-based textual similarity for visual sign detection.
- Applied to 87,668 tablet fragments from the eBL corpus, the method produces nearly 2.9 million sign detections.
- The approach operates without linguistic priors but is sensitive to tablet damage and layout variability, providing a scalable foundation for cuneiform analysis.
Show HN: EPEE, an expert-annotated ASL dataset from native Deaf signers
2 months ago
- CLERC Épée v0.2 is an AI-grade sign language dataset for training, benchmarking, and inter-signer variability studies.
- It contains 600 ASL clips (150 unique phrases) signed by four Deaf signers in a fully parallel structure.
- Data includes extracted keypoints (hands, body, eyes, mouth, head silhouette) and gloss annotations.
- A benchmark shows training on multiple signers improves generalization to new signers.
- Keypoints are in MediaPipe image-normalized space with 128 landmarks per frame.
- The dataset is released under CC BY-NC-SA 4.0, with commercial licensing available.
- Intended for research and educational use, not for surveillance or biometric applications.

dataset

How to easily dig through the Epstein files yourself

Automated sign detection across the Electronic Babylonian Library

Show HN: EPEE, an expert-annotated ASL dataset from native Deaf signers