Link Search Menu Expand Document

2D Classification

Track 1: In-domain classification (Scratch)

The model is trained and tested on PeRFception-Co3D dataset. We sort models with the primary metric Acc@1. For this track, the models trained from scratch is allowed. No external dataset is allowed. Acc@1 stands for top-1 accuracy, and Acc@5 stands for top-5 accuracy.

ModelAcc@1Acc@5CheckpointsCode
ResNext10185.48 \(\pm\) 0.0696.26 \(\pm\) 0.03linklink
WideResNet10185.30 \(\pm\) 0.1196.31 \(\pm\) 0.10linklink
ResNet15285.28 \(\pm\) 0.0296.39 \(\pm\) 0.06linklink
ResNet10185.11 \(\pm\) 0.2396.32 \(\pm\) 0.12linklink
WideResNet5084.68 \(\pm\) 0.0296.03 \(\pm\) 0.03linklink
ResNext5084.32 \(\pm\) 0.1895.92 \(\pm\) 0.16linklink
ResNet5083.77 \(\pm\) 0.0895.99 \(\pm\) 0.08linklink
ResNet3483.61 \(\pm\) 0.0495.89 \(\pm\) 0.06linklink
ResNet1882.05 \(\pm\) 0.2495.37 \(\pm\) 0.05linklink

Track 2: In-domain classification (Pre-trained)

The model is trained on PeRFception-Co3D dataset and tested CO3D dataset. We sort models with the primary metric Acc@1. For this track, any pre-trained datasets are allowed. However, we prohibit external datasets that include any CO3D test images on training split.

ModelDatasetAcc@1Acc@5CheckpointsCode
ResNet152ImageNet88.73 \(\pm\) 0.1597.24 \(\pm\) 0.08linklink
WideResNet101ImageNet88.39 \(\pm\) 0.0796.31 \(\pm\) 0.10linklink
ResNext101ImageNet88.51 \(\pm\) 0.1696.93 \(\pm\) 0.07linklink
ResNet101ImageNet88.32 \(\pm\) 0.1397.13 \(\pm\) 0.05linklink
WideResNet50ImageNet87.75 \(\pm\) 0.2596.84 \(\pm\) 0.09linklink
ResNext50ImageNet87.30 \(\pm\) 0.1696.66 \(\pm\) 0.07linklink
ResNet50ImageNet87.30 \(\pm\) 0.0896.69 \(\pm\) 0.08linklink
ResNet34ImageNet86.25 \(\pm\) 0.1996.50 \(\pm\) 0.09linklink
ResNet18ImageNet84.97 \(\pm\) 0.1396.24 \(\pm\) 0.09linklink