Submission details of DETR

Name DETR
Paper Link https://arxiv.org/abs/2005.12872
Code Link https://github.com/facebookresearch/detr
AP 0.48664
Input N/A
Runtime 0.35 s
Environment 1 GPU (GTX 1060)
Abstract We present a new method that views object detection as a direct set prediction problem. Our approach streamlines the detection pipeline, effectively removing the need for many hand-designed components like a non-maximum suppression procedure or anchor generation that explicitly encode our prior knowledge about the task. The main ingredients of the new framework, called DEtection TRansformer or DETR, are a set-based global loss that forces unique predictions via bipartite matching, and a transformer encoder-decoder architecture. Given a fixed small set of learned object queries, DETR reasons about the relations of the objects and the global image context to directly output the final set of predictions in parallel. The new model is conceptually simple and does not require a specialized library, unlike many other modern detectors. DETR demonstrates accuracy and run-time performance on par with the well-established and highly-optimized Faster R-CNN baseline on the challenging COCO object detection dataset. Moreover, DETR can be easily generalized to produce panoptic segmentation in a unified manner. We show that it significantly outperforms competitive baselines. Training code and pretrained models are available at https://github.com/facebookresearch/detr.

Detailed results

Overall precision/recall curve

Per-sequence results

sequence name AP
cubberly-auditorium-2019-04-22_1 0.614066
discovery-walk-2019-02-28_0 0.549403
discovery-walk-2019-02-28_1 0.576734
food-trucks-2019-02-12_0 0.597451
gates-ai-lab-2019-04-17_0 0.684306
gates-basement-elevators-2019-01-17_0 0.690566
gates-foyer-2019-01-17_0 0.747334
gates-to-clark-2019-02-28_0 0.636902
hewlett-class-2019-01-23_0 0.788844
hewlett-class-2019-01-23_1 0.885695
huang-2-2019-01-25_1 0.497532
huang-intersection-2019-01-22_0 0.456637
indoor-coupa-cafe-2019-02-06_0 0.478063
lomita-serra-intersection-2019-01-30_0 0.383418
meyer-green-2019-03-16_1 0.313803
nvidia-aud-2019-01-25_0 0.416606
nvidia-aud-2019-04-18_1 0.604265
nvidia-aud-2019-04-18_2 0.630602
outdoor-coupa-cafe-2019-02-06_0 0.397808
quarry-road-2019-02-28_0 0.413082
serra-street-2019-01-30_0 0.358779
stlc-111-2019-04-19_1 0.759333
stlc-111-2019-04-19_2 0.742592
tressider-2019-03-16_2 0.458202
tressider-2019-04-26_0 0.384417
tressider-2019-04-26_1 0.476278
tressider-2019-04-26_3 0.423914
total 0.48664