Recent high-performing object detectors, such asFaster R-CNN [34],formulate thedetection problem as atwo-stage pipeline.

Kinetics action classification.Our settings mainly follow[31,77]. The settings mainly follow [39, 77]. We report top-1 and top-5accuracyonthevalidation set. Entriesusingspatialresolution >2242 are noted in gray; entries using in-house data for supervision are in light blue. Importantly, our method is muchsimpler than many other entries.

We observe that MAE can outperform supervised pre-training by large margins.

As aresult, there is aproliferation of distinct architectures and loss functions for different vision tasks.

UOTA adaptively searches for the most important sampling region to produce views, and provides viable choice for outlier-robust self-supervised learning approaches.

We follow [8] to train the depth estimation model and compute thedepth estimation error onpixelsassociated with ground truth LiDARpoints.

Somelaterapproaches [40, 28, 23, 22, 38, 43] try to explicitly model this relationship in a learnable fashion.

A.1 NASSearchSpaces NASBench-1011 introduces a large and expressive search space with 423k unique convolutional neural architectures and training statistics on CIFAR-10. NASBench-2012 contains the training statistics of15,625 architectures across three different datasets, including CIFAR-10, CIFAR-100, and Tiny-ImageNet-16. NASBench-NLP5 [10] is an NLP neural architecture search space, including 14k recurrent cells trained on the Penn Treebank (PTB)[11]dataset. Thegenerated noise maps aredirectly multiplied bythelevelwhich canbeselected from 0 to 1 with a step of 0.1. Other settings are already described in Section 3.1.2.

It explores two essential properties, namelydataset-level uniqueness and transformation equivariance, of the relation between queries and instances.

Semi-supervised learning aims to leverage alarge amount of unlabeled data for performance boosting. Existing works primarily focus on image classification. Inthispaper,wedelveintosemi-supervised learning forobject detection, where labeled data are more labor-intensive to collect.