We present an interesting and challenging dataset that features a large number of scenes with messy tables captured from multiple camera views. Each scene in this dataset is highly complex, containing multiple object instances that could be identical, stacked and occluded by other instances. The key challenge is to associate all instances given the RGB image of all views. The seemingly simple task surprisingly fails many popular methods or heuristics that we assume good performance in object association. The dataset challenges existing methods in mining subtle appearance differences, reasoning based on contexts, and fusing appearance with geometric cues for establishing an association. We report interesting findings with some popular baselines, and discuss how this dataset could help inspire new problems and catalyse more robust formulations to tackle real-world instance association problems.

	[Preprint]	[Code]

	Aliyun	Google Drive

Baselines Model/Method AP FPR-95 IPAA-100 IPAA-90 IPAA-80 Homography 0.049 0.944 0 0 0 SIFT 0.063 0.866 0 0 0 MatchNet 0.193 0.458 0.010 0.012 0.033 MatchNet(ResNet-18) 0.138 0.410 0.002 0.003 0.010 DeepCompare 0.202 0.412 0.023 0.025 0.063 DeepCompare(ResNet-18) 0.129 0.402 0.005 0.005 0.010 DeepDesc 0.090 0.906 0.011 0.011 0.018 DeepDesc(ResNet-18) 0.171 0.804 0.027 0.032 0.058 TripletNet 0.467 0.206 0.168 0.220 0.376 TripletNet+Zoom Out 0.430 0.269 0.047 0.062 0.161 ASNet 0.524 0.209 0.170 0.241 0.418 ASNet+Epipolar Soft Constraint 0.577 0.157 0.219 0.306 0.499 Statistics Classes Cameras Setups Relative Poses Scenes Images BBoxes Instance per Scene 120 9 567 20,412 5,579 50,211 1,219,240 6-73 Full List of Objects Acknowledgements This research was supported by SenseTime-NTU Collaboration Project, Singapore MOE AcRF Tier 1 (2018-T1-002-056), NTU SUG, and NTU NAP Total Visitor Count:       Unique Visitor Count: