Introduction
license: apache-2.0
language: en
BUDDI Table Factory: A toolbox for generating synthetic documents with annotated tables and cells
About
In Cell detection, we initialize the weights with a pre-trained CDeCNet model using COCO dataset. We re-train the model for five epochs using a stochastic gradient descent optimizer with a learning rate of 0.00125, the momentum of 0.9, and weight decay of 0.0001.
Hardware Used
We perform all the experiments on NVIDIA GeForce RTX 2080 Ti GPU with 12 GB GPU memory, Intel(R) Xeon(R) CPU E5-2640 v2 @ 2.00GHz, and 128 GB of RAM.
Table Detection Model & Training Parameter
Optimizer
| Parameter | Value |
|---|
| Type | SGD |
| Learning Rate | 0.00125 |
| Momentum | 0.8 |
| Weight Decay | 0.001 |
*** Learning Policy ***
| Parameter | Value |
|---|
| Policy | Step |
| Warmup | Linear |
| Warmup Iteration | 100 |
| Warmup Ratio | 0.001 |
| Step | 4,16,32 |
General Parameter
| Parameter | Value |
|---|
| Epoch | 10 |
| Step Interval | 50 |
Model Paper Reference
CDeC-Net: Composite Deformable Cascade Network for Table Detection in Document Images
Citation
If you find BTF useful for your work, please cite the following paper:
