Submission details of RetinaNet

Name RetinaNet
Paper Link https://arxiv.org/abs/1708.02002
Code Link https://github.com/facebookresearch/detectron2
AP 0.503799
Input N/A
Runtime 0.056 s
Environment 1 GPU (Titan X)
Abstract The highest accuracy object detectors to date are based on a two-stage approach popularized by R-CNN, where a classifier is applied to a sparse set of candidate object locations. In contrast, one-stage detectors that are applied over a regular, dense sampling of possible object locations have the potential to be faster and simpler, but have trailed the accuracy of two-stage detectors thus far. In this paper, we investigate why this is the case. We discover that the extreme foreground-background class imbalance encountered during training of dense detectors is the central cause. We propose to address this class imbalance by reshaping the standard cross entropy loss such that it down-weights the loss assigned to well-classified examples. Our novel Focal Loss focuses training on a sparse set of hard examples and prevents the vast number of easy negatives from overwhelming the detector during training. To evaluate the effectiveness of our loss, we design and train a simple dense detector we call RetinaNet. Our results show that when trained with the focal loss, RetinaNet is able to match the speed of previous one-stage detectors while surpassing the accuracy of all existing state-of-the-art two-stage detectors.

Detailed results

Overall precision/recall curve

Per-sequence results

sequence name AP
cubberly-auditorium-2019-04-22_1 0.532919
discovery-walk-2019-02-28_0 0.482166
discovery-walk-2019-02-28_1 0.562439
food-trucks-2019-02-12_0 0.559675
gates-ai-lab-2019-04-17_0 0.668771
gates-basement-elevators-2019-01-17_0 0.765753
gates-foyer-2019-01-17_0 0.798636
gates-to-clark-2019-02-28_0 0.626826
hewlett-class-2019-01-23_0 0.867171
hewlett-class-2019-01-23_1 0.874166
huang-2-2019-01-25_1 0.482307
huang-intersection-2019-01-22_0 0.455224
indoor-coupa-cafe-2019-02-06_0 0.509853
lomita-serra-intersection-2019-01-30_0 0.420568
meyer-green-2019-03-16_1 0.308499
nvidia-aud-2019-01-25_0 0.425777
nvidia-aud-2019-04-18_1 0.612574
nvidia-aud-2019-04-18_2 0.557414
outdoor-coupa-cafe-2019-02-06_0 0.34839
quarry-road-2019-02-28_0 0.398908
serra-street-2019-01-30_0 0.321111
stlc-111-2019-04-19_1 0.755117
stlc-111-2019-04-19_2 0.733923
tressider-2019-03-16_2 0.366394
tressider-2019-04-26_0 0.433025
tressider-2019-04-26_1 0.529644
tressider-2019-04-26_3 0.498012
total 0.503799