Table Transformer (TATR) - PubTables1M Table Structure Recognition is a Transformer-based object detection model trained on the PubTables1M dataset. It is designed for detecting and identifying table structures, such as rows, columns, and table boundaries, in unstructured scientific and academic documents. Built on the DETR (DEtection TRansformer) framework, the model applies a "normalize before" approach, where layer normalization occurs before self- and cross-attention mechanisms. This model is particularly useful for document digitization, research paper analysis, and structured data extraction from complex table layouts in PDFs and scanned images.
MIT
Sector Agnostic
Microsoft
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
Open
12/03/25 06:34:59
N.A.
0
© 2026 - Copyright AIKosh. All rights reserved.