super-resolution-detection

End-to-End Super Resolution Object Detection Networks

Models

https://drive.google.com/drive/u/1/folders/1qXuCJbsDp7Hno4JUwanviMIWDJG1-eck

Progress

Data Set VOC0712

• Training: VOC12 (17128 images)
• mAP Test: VOC07 ( 4952 images)

DBPN: Original DBPN-4x model (DIV2K pretrained). Used for 3rd column, and as initial weight for 4th

• SSD300 mAP

300x300 HR	300x300 LR (75x75 Bicubic Upscale)	75x75 LR => DBPNx4	75x75 LR => SROD (4000 end-2-end iters)
76.99%	48.65%	39.84%	41.42%

DBPN: Retrained (50 Epochs / VOC12) DBPN-4x model (start with DIV2K pretrained). Used for 3rd column, and as initial weight for 4th

• SSD300 mAP

300x300 HR	300x300 LR (75x75 Bicubic Upscale)	75x75 LR => DBPNx4	75x75 LR => SROD (pending end-2-end )
76.99%	48.65%	48.52%

New input Images scaled by imageScale300.ipynb

DBPN: Retrained (50 Epochs / VOC12) DBPN-4x model (start with DIV2K pretrained). Used for 5th column, and as initial weight for 6th

• SSD300 mAP

300x300 HR	300x300 LR (75x75 Bicubic)	300x300 LR (75x75 Bilinear)	300x300 LR (75x75 Nearest)	75x75 LR => DBPNx4	75x75 LR => SROD (end-2-end )
77.34%	48.91%	46.92%	7.17%	49.59%	TBD

DBPN: Intensively Retrained (100 Epochs / VOC12) DBPN-4x base model (start with DIV2K pretrained). Used for 5th column, and as SROD's initial S-Net weight for 6th

• SSD300 mAP

300x300 HR	300x300 LR (75x75 Bicubic)	300x300 LR (75x75 Bilinear)	75x75 LR (in 300x300 Bgrd)	300x300 LR (75x75 Nearest)	75x75 LR => DBPNx4	75x75 LR => SROD (e2e Retrain 16 Epochs)
77.34%	48.91%	46.92%	40.35%	7.17%	56.08%	62.89%

Authorship

This project is equally contributed by Hai Xiao and Kegang Xu, with extra code components from following authors:

Citation

@article{SROD2018,
    Author = {Hai Xiao and Kegang Xu},
    Title = {A Pytorch Implementation of Super Resolution Object Detection},
    Journal = {https://github.com/cs231x/super-resolution-detection},
    Year = {2018}
}

@inproceedings{DBPN2018,
    title={Deep Back-Projection Networks for Super-Resolution},
    author={Haris, Muhammad and Shakhnarovich, Greg and Ukita, Norimichi},
    booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
    year={2018}
}

@article{SSD300,
    Author = {Max deGroot and Ellis Brown},
    Title = {SSD: Single Shot MultiBox Object Detector, in PyTorch},
    Journal = {https://github.com/amdegroot/ssd.pytorch},
    Year = {2017}
}