Writing the Report: - Githubissues

m-kashani commented 4 years ago

Detectors for Object detection:

You Only Look Once (YOLO) YOLO works at 30FPS and said to have ~57% MAP on COCO dataset. Paper: 30 Mar 2016 -1 https://pjreddie.com/media/files/papers/yolo.pdf -2 https://arxiv.org/pdf/1612.08242.pdf
YOLOv3 various improvements to the YOLO detection method. Better? Faster? Stronger? paper 25 Dec 2016: https://arxiv.org/pdf/1612.08242.pdf
Faster - RCNN it's said to be slower than YOLO and SSD, but if one has time and resources it can be tried. Regional Proposal Network (RPN) is proposed. RPN is proposed to generate high-quality region proposals. RPN simultaneously predicts object bounds and objectness scores at each position. Paper 6 Jan 2016: https://arxiv.org/pdf/1506.01497.pdf
R-FCN: previous region-based detectors such as Fast/Faster R-CNN apply a costly per-region subnetwork hundreds of times. region-based, fully convolutional networks for accurate and efficient object detection. ?Multi-scale training and testing are used on some results Paper 21 Jun 2016: https://arxiv.org/pdf/1605.06409.pdf
Single Shot Multibox Detector (SSD): 100x faster than YOLO and faster RCNN (they claim in paper). (SSD300 and SSD512 applies data augmentation for small objects to improve mAP.) Paper: -1 https://storage.googleapis.com/pub-tools-public-publication-data/pdf/44872.pdf -2 https://arxiv.org/pdf/1512.02325.pdf
FPN (Feature Pyramid Networks for Object Detection)
paper 19 April 2017: https://arxiv.org/pdf/1612.03144.pdf
ًRetinaNET (Focal Loss for Dense Object Detection)
paper 7 Feb 2018: https://arxiv.org/pdf/1708.02002.pdf

Comparing Paper Results: It is unwise to compare results side-by-side from different papers. Those experiments are done in different settings which are not purposed for apple-to-apple comparisons.

Evaluation:

http://cocodataset.org/#detection-eval