In the real open world, data tends to follow long-tailed class distributions, motivating the well-studied long-tailed recognition (LTR) problem. Naive training produces models that are biased toward common classes in terms of higher accuracy. The key to addressing LTR is to balance various aspects including data distribution, training losses, and gradients in learning. We explore an orthogonal direction, weight balancing, motivated by the empirical observation that the naively trained classifier has "artificially" larger weights in norm for common classes (because there exists abundant data to train them, unlike the rare classes). We investigate three techniques to balance weights, L2-normalization, weight decay, and MaxNorm. We first point out that L2-normalization "perfectly" balances per-class weights to be unit norm, but such a hard constraint might prevent classes from learning better classifiers. In contrast, weight decay penalizes larger weights more heavily and so learns small balanced weights; the MaxNorm constraint encourages growing small weights within a norm ball but caps all the weights by the radius. Our extensive study shows that both help learn balanced weights and greatly improve the LTR accuracy. Surprisingly, weight decay, although underexplored in LTR, significantly improves over prior work. Therefore, we adopt a two-stage training paradigm and propose a simple approach to LTR: (1) learning features using the cross-entropy loss by tuning weight decay, and (2) learning classifiers using class-balanced loss by tuning weight decay and MaxNorm. Our approach achieves the state-of-the-art accuracy on five standard benchmarks, serving as a future baseline for long-tailed recognition.
Continual learning (CL) is widely regarded as crucial challenge for lifelong AI. However, existing CL benchmarks, e.g. Permuted-MNIST and Split-CIFAR, make use of artificial temporal variation and do not align with or generalize to the real-world. In this paper, we introduce CLEAR, the first continual image classification benchmark dataset with a natural temporal evolution of visual concepts in the real world that spans a decade (2004-2014). We build CLEAR from existing large-scale image collections (YFCC100M) through a novel and scalable low-cost approach to visio-linguistic dataset curation. Our pipeline makes use of pretrained vision-language models (e.g. CLIP) to interactively build labeled datasets, which are further validated with crowd-sourcing to remove errors and even inappropriate images (hidden in original YFCC100M). The major strength of CLEAR over prior CL benchmarks is the smooth temporal evolution of visual concepts with real-world imagery, including both high-quality labeled data along with abundant unlabeled samples per time period for continual semi-supervised learning. We find that a simple unsupervised pre-training step can already boost state-of-the-art CL algorithms that only utilize fully-supervised data. Our analysis also reveals that mainstream CL evaluation protocols that train and test on iid data artificially inflate performance of CL system. To address this, we propose novel "streaming" protocols for CL that always test on the (near) future. Interestingly, streaming protocols (a) can simplify dataset curation since today's testset can be repurposed for tomorrow's trainset and (b) can produce more generalizable models with more accurate estimates of performance since all labeled data from each time-period is used for both training and testing (unlike classic iid train-test splits).
Prior work for articulated 3D shape reconstruction often relies on specialized sensors (e.g., synchronized multi-camera systems), or pre-built 3D deformable models (e.g., SMAL or SMPL). Such methods are not able to scale to diverse sets of objects in the wild. We present BANMo, a method that requires neither a specialized sensor nor a pre-defined template shape. BANMo builds high-fidelity, articulated 3D models (including shape and animatable skinning weights) from many monocular casual videos in a differentiable rendering framework. While the use of many videos provides more coverage of camera views and object articulations, they introduce significant challenges in establishing correspondence across scenes with different backgrounds, illumination conditions, etc. Our key insight is to merge three schools of thought; (1) classic deformable shape models that make use of articulated bones and blend skinning, (2) volumetric neural radiance fields (NeRFs) that are amenable to gradient-based optimization, and (3) canonical embeddings that generate correspondences between pixels and an articulated model. We introduce neural blend skinning models that allow for differentiable and invertible articulated deformations. When combined with canonical embeddings, such models allow us to establish dense correspondences across videos that can be self-supervised with cycle consistency. On real and synthetic datasets, BANMo shows higher-fidelity 3D reconstructions than prior works for humans and animals, with the ability to render realistic images from novel viewpoints and poses. Project webpage: banmo-www.github.io .
We explore how to leverage neural radiance fields (NeRFs) to build interactive 3D environments from large-scale visual captures spanning buildings or even multiple city blocks collected primarily from drone data. In contrast to the single object scenes against which NeRFs have been traditionally evaluated, this setting poses multiple challenges including (1) the need to incorporate thousands of images with varying lighting conditions, all of which capture only a small subset of the scene, (2) prohibitively high model capacity and ray sampling requirements beyond what can be naively trained on a single GPU, and (3) an arbitrarily large number of possible viewpoints that make it unfeasible to precompute all relevant information beforehand (as real-time NeRF renderers typically do). To address these challenges, we begin by analyzing visibility statistics for large-scale scenes, motivating a sparse network structure where parameters are specialized to different regions of the scene. We introduce a simple geometric clustering algorithm that partitions training images (or rather pixels) into different NeRF submodules that can be trained in parallel. We evaluate our approach across scenes taken from the Quad 6k and UrbanScene3D datasets as well as against our own drone footage and show a 3x training speedup while improving PSNR by over 11% on average. We subsequently perform an empirical evaluation of recent NeRF fast renderers on top of Mega-NeRF and introduce a novel method that exploits temporal coherence. Our technique achieves a 40x speedup over conventional NeRF rendering while remaining within 0.5 db in PSNR quality, exceeding the fidelity of existing fast renderers.
Existing state-of-the-art methods for Video Object Segmentation (VOS) learn low-level pixel-to-pixel correspondences between frames to propagate object masks across video. This requires a large amount of densely annotated video data, which is costly to annotate, and largely redundant since frames within a video are highly correlated. In light of this, we propose HODOR: a novel method that tackles VOS by effectively leveraging annotated static images for understanding object appearance and scene context. We encode object instances and scene information from an image frame into robust high-level descriptors which can then be used to re-segment those objects in different frames. As a result, HODOR achieves state-of-the-art performance on the DAVIS and YouTube-VOS benchmarks compared to existing methods trained without video annotations. Without any architectural modification, HODOR can also learn from video context around single annotated video frames by utilizing cyclic consistency, whereas other methods rely on dense, temporally consistent annotations.
Recent history has seen a tremendous growth of work exploring implicit representations of geometry and radiance, popularized through Neural Radiance Fields (NeRF). Such works are fundamentally based on a (implicit) volumetric representation of occupancy, allowing them to model diverse scene structure including translucent objects and atmospheric obscurants. But because the vast majority of real-world scenes are composed of well-defined surfaces, we introduce a surface analog of such implicit models called Neural Reflectance Surfaces (NeRS). NeRS learns a neural shape representation of a closed surface that is diffeomorphic to a sphere, guaranteeing water-tight reconstructions. Even more importantly, surface parameterizations allow NeRS to learn (neural) bidirectional surface reflectance functions (BRDFs) that factorize view-dependent appearance into environmental illumination, diffuse color (albedo), and specular "shininess." Finally, rather than illustrating our results on synthetic scenes or controlled in-the-lab capture, we assemble a novel dataset of multi-view images from online marketplaces for selling goods. Such "in-the-wild" multi-view image sets pose a number of challenges, including a small number of views with unknown/rough camera estimates. We demonstrate that surface-based neural reconstructions enable learning from such data, outperforming volumetric neural rendering-based reconstructions. We hope that NeRS serves as a first step toward building scalable, high-quality libraries of real-world shape, materials, and illumination. The project page with code and video visualizations can be found at https://jasonyzhang.com/ners.
For machine learning models trained with limited labeled training data, validation stands to become the main bottleneck to reducing overall annotation costs. We propose a statistical validation algorithm that accurately estimates the F-score of binary classifiers for rare categories, where finding relevant examples to evaluate on is particularly challenging. Our key insight is that simultaneous calibration and importance sampling enables accurate estimates even in the low-sample regime (< 300 samples). Critically, we also derive an accurate single-trial estimator of the variance of our method and demonstrate that this estimator is empirically accurate at low sample counts, enabling a practitioner to know how well they can trust a given low-sample estimate. When validating state-of-the-art semi-supervised models on ImageNet and iNaturalist2017, our method achieves the same estimates of model performance with up to 10x fewer labels than competing approaches. In particular, we can estimate model F1 scores with a variance of 0.005 using as few as 100 labels.
Efficient processing of high-resolution video streams is safety-critical for many robotics applications such as autonomous driving. Image downsampling is a commonly adopted technique to ensure the latency constraint is met. However, this naive approach greatly restricts an object detector's capability to identify small objects. In this paper, we propose an attentional approach that elastically magnifies certain regions while maintaining a small input canvas. The magnified regions are those that are believed to have a high probability of containing an object, whose signal can come from a dataset-wide prior or frame-level prior computed from recent object predictions. The magnification is implemented by a KDE-based mapping to transform the bounding boxes into warping parameters, which are then fed into an image sampler with anti-cropping regularization. The detector is then fed with the warped image and we apply a differentiable backward mapping to get bounding box outputs in the original space. Our regional magnification allows algorithms to make better use of high-resolution input without incurring the cost of high-resolution processing. On the autonomous driving datasets Argoverse-HD and BDD100K, we show our proposed method boosts the detection AP over standard Faster R-CNN, with and without finetuning. Additionally, building on top of the previous state-of-the-art in streaming detection, our method sets a new record for streaming AP on Argoverse-HD (from 17.8 to 23.0 on a GTX 1080 Ti GPU), suggesting that it has achieved a superior accuracy-latency tradeoff.
One common failure mode of Neural Radiance Field (NeRF) models is fitting incorrect geometries when given an insufficient number of input views. We propose DS-NeRF (Depth-supervised Neural Radiance Fields), a loss for learning neural radiance fields that takes advantage of readily-available depth supervision. Our key insight is that sparse depth supervision can be used to regularize the learned geometry, a crucial component for effectively rendering novel views using NeRF. We exploit the fact that current NeRF pipelines require images with known camera poses that are typically estimated by running structure-from-motion (SFM). Crucially, SFM also produces sparse 3D points that can be used as ``free" depth supervision during training: we simply add a loss to ensure that depth rendered along rays that intersect these 3D points is close to the observed depth. We find that DS-NeRF can render more accurate images given fewer training views while training 2-6x faster. With only two training views on real-world images, DS-NeRF significantly outperforms NeRF as well as other sparse-view variants. We show that our loss is compatible with these NeRF models, demonstrating that depth is a cheap and easily digestible supervisory signal. Finally, we show that DS-NeRF supports other types of depth supervision such as scanned depth sensors and RGBD reconstruction outputs.
Remarkable progress has been made in 3D reconstruction of rigid structures from a video or a collection of images. However, it is still challenging to reconstruct nonrigid structures from RGB inputs, due to its under-constrained nature. While template-based approaches, such as parametric shape models, have achieved great success in modeling the "closed world" of known object categories, they cannot well handle the "open-world" of novel object categories or outlier shapes. In this work, we introduce a template-free approach to learn 3D shapes from a single video. It adopts an analysis-by-synthesis strategy that forward-renders object silhouette, optical flow, and pixel values to compare with video observations, which generates gradients to adjust the camera, shape and motion parameters. Without using a category-specific shape template, our method faithfully reconstructs nonrigid 3D structures from videos of human, animals, and objects of unknown classes. Code will be available at lasr-google.github.io .