Rethinking pose estimation in crowds: overcoming the detection information-bottleneck and ambiguity

Benchmark Model Rank Results
animal-pose-estimation-on-fish-100HRNet-W48 + Faster R-CNN#1mAP: 89.1
animal-pose-estimation-on-fish-100BUCTD-preNet-W48 (DLCRNet)#2mAP: 88.7
animal-pose-estimation-on-fish-100BUCTD-preNet-W48 (CID-W32)#3mAP: 88.0
animal-pose-estimation-on-marmoset-8kBUCTD-preNet-W48 (CID-W32)#1mAP: 93.3
animal-pose-estimation-on-marmoset-8kCID-W32#2mAP: 92.5
animal-pose-estimation-on-marmoset-8kBUCTD-CoAM-W48 (DLCRNet)#3mAP: 91.6
animal-pose-estimation-on-trimouse-161BUCTD-CoAM-W48 (DLCRNet)#1mAP: 99.1
animal-pose-estimation-on-trimouse-161DLCRNet#3mAP: 95.8
animal-pose-estimation-on-trimouse-161CID-W32#6mAP: 86.8
multi-person-pose-estimation-on-crowdposeBUCTD-W48 (w/cond. input from PETR, and generative sampling)#2mAP @0.5:0.95: 78.5AP Easy: 83.9AP Medium: 79.0AP Hard: 72.3
pose-estimation-on-cocoBUCTD (PETR, with generative sampling)#4AP: 77.8
pose-estimation-on-cocoBUCTD (PETR, with generative sampling)#10APL: 83.7APM: 74.2
pose-estimation-on-crowdposeBUCTD-W48 (w/cond. input from PETR, and generative sampling)#1AP: 78.5AP Hard: 72.3AP Easy: 83.9AP Medium: 79.0
pose-estimation-on-crowdposeBUCTD-W48 (w/cond. input from PETR)#3AP: 76.7
pose-estimation-on-crowdposeBUCTD-W48#6AP: 72.9
pose-estimation-on-ochumanBUCTD (CID-W32)#6Test AP: 47.2Validation AP: 47.7