D-FINE & DEIM Pretrained Models and Dataset Preparation

This repository provides pretrained models for D-FINE and DEIM, along with datasets and conversion scripts for training and evaluation. It includes:

• VOC and COCO-format datasets
• Pretrained model weights
• Utility script for data restructuring

Contents

📂 Datasets

• PASCAL-VOC-2007-1/: Original dataset in Pascal VOC format.
• PASCAL_VOC_2007_COCO_Format/: Dataset converted to COCO format for compatibility with various deep learning frameworks.

📜 Scripts

• convert_restructure.ipynb: A Jupyter notebook for converting and restructuring the dataset formats as needed (e.g., Pascal VOC to COCO).

🧠 Pretrained Models

• deim_dfine_hgnetv2_s_coco_120e.pth: Pretrained model for D-FINE/DEIM using HGNetV2 architecture, trained on COCO for 120 epochs.
• dfine_s_coco.pth: Pretrained D-FINE model on COCO.

Getting Started

1. Clone this repository

   git clone https://github.com/nekloyh/VOC_2007_in_Coco_format.git
   cd VOC_2007_in_Coco_format

2. Download or Prepare Data

Ensure the datasets are in the correct format. You can use the convert_restructure.ipynb notebook to convert between formats if needed.