https://github.com/lessw2020/training-detr

Unofficial Colab on how to train DETR, the intelligent object detector, with your own dataset. DETR = Detection Transformer
https://github.com/lessw2020/training-detr

Last synced: about 2 months ago
JSON representation

Unofficial Colab on how to train DETR, the intelligent object detector, with your own dataset. DETR = Detection Transformer

• Host: GitHub
• URL: https://github.com/lessw2020/training-detr
• Owner: lessw2020
• License: apache-2.0
• Created: 2020-06-13T18:51:05.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2020-06-29T01:55:07.000Z (about 4 years ago)
• Last Synced: 2024-07-09T20:18:44.702Z (3 months ago)
• Language: Jupyter Notebook
• Homepage:
• Size: 52.7 KB
• Stars: 40
• Watchers: 6
• Forks: 6
• Open Issues: 2
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# training-detr

Unofficial Colab on how to train DETR, the intelligent object detector, with your own dataset.  DETR = Detection Transformer

Note there are two methods for training DETR:

1 - Fine tune the existing head.  This is the initial process I will setup here.

2 - Train from scratch.  This is the next process I'll show.

Generally fine tuning the pre-trained head is likely the best place to start. 

Note that your dataset must be in COCO format.  

Updates - Detr is now available in detectron2 so that may be an option for some.  

I've added a custom dataset class which will handle the class id remapping needed to train with DETR directly.

DETR was developed and open sourced by Facebook AI Research - link to their github which should supercede any info here when in doubt:

https://github.com/facebookresearch/detr