Jose M. Álvarez

Yilun Chen, Zhiding Yu, Yukang Chen et al.

False negatives (FN) in 3D object detection, {\em e.g.}, missing predictions of pedestrians, vehicles, or other obstacles, can lead to potentially dangerous situations in autonomous driving. While being fatal, this issue is understudied in many current 3D detection methods. In this work, we propose Hard Instance Probing (HIP), a general pipeline that identifies \textit{FN} in a multi-stage manner and guides the models to focus on excavating difficult instances. For 3D object detection, we instantiate this method as FocalFormer3D, a simple yet effective detector that excels at excavating difficult objects and improving prediction recall. FocalFormer3D features a multi-stage query generation to discover hard objects and a box-level transformer decoder to efficiently distinguish objects from massive object candidates. Experimental results on the nuScenes and Waymo datasets validate the superior performance of FocalFormer3D. The advantage leads to strong performance on both detection and tracking, in both LiDAR and multi-modal settings. Notably, FocalFormer3D achieves a 70.5 mAP and 73.9 NDS on nuScenes detection benchmark, while the nuScenes tracking benchmark shows 72.1 AMOTA, both ranking 1st place on the nuScenes LiDAR leaderboard. Our code is available at \url{https://github.com/NVlabs/FocalFormer3D}.

Zhiqi Li, Zhiding Yu, Wenhai Wang et al.

View Transformation Module (VTM), where transformations happen between multi-view image features and Bird-Eye-View (BEV) representation, is a crucial step in camera-based BEV perception systems. Currently, the two most prominent VTM paradigms are forward projection and backward projection. Forward projection, represented by Lift-Splat-Shoot, leads to sparsely projected BEV features without post-processing. Backward projection, with BEVFormer being an example, tends to generate false-positive BEV features from incorrect projections due to the lack of utilization on depth. To address the above limitations, we propose a novel forward-backward view transformation module. Our approach compensates for the deficiencies in both existing methods, allowing them to enhance each other to obtain higher quality BEV representations mutually. We instantiate the proposed module with FB-BEV, which achieves a new state-of-the-art result of 62.4% NDS on the nuScenes test set. Code and models are available at https://github.com/NVlabs/FB-BEV.

Ali Hatamizadeh, Greg Heinrich, Hongxu Yin et al.

We design a new family of hybrid CNN-ViT neural networks, named FasterViT, with a focus on high image throughput for computer vision (CV) applications. FasterViT combines the benefits of fast local representation learning in CNNs and global modeling properties in ViT. Our newly introduced Hierarchical Attention (HAT) approach decomposes global self-attention with quadratic complexity into a multi-level attention with reduced computational costs. We benefit from efficient window-based self-attention. Each window has access to dedicated carrier tokens that participate in local and global representation learning. At a high level, global self-attentions enable the efficient cross-window communication at lower costs. FasterViT achieves a SOTA Pareto-front in terms of accuracy and image throughput. We have extensively validated its effectiveness on various CV tasks including classification, object detection and segmentation. We also show that HAT can be used as a plug-and-play module for existing networks and enhance them. We further demonstrate significantly faster and more accurate performance than competitive counterparts for images with high resolution. Code is available at https://github.com/NVlabs/FasterViT.

Rafid Mahmood, James Lucas, Jose M. Alvarez et al. · utoronto

Modern deep learning systems require huge data sets to achieve impressive performance, but there is little guidance on how much or what kind of data to collect. Over-collecting data incurs unnecessary present costs, while under-collecting may incur future costs and delay workflows. We propose a new paradigm for modeling the data collection workflow as a formal optimal data collection problem that allows designers to specify performance targets, collection costs, a time horizon, and penalties for failing to meet the targets. Additionally, this formulation generalizes to tasks requiring multiple data sources, such as labeled and unlabeled data used in semi-supervised learning. To solve our problem, we develop Learn-Optimize-Collect (LOC), which minimizes expected future collection costs. Finally, we numerically compare our framework to the conventional baseline of estimating data requirements by extrapolating from neural scaling laws. We significantly reduce the risks of failing to meet desired performance targets on several classification, segmentation, and detection tasks, while maintaining low total collection costs.

Lingchen Meng, Shiyi Lan, Hengduo Li et al.

In-context segmentation aims at segmenting novel images using a few labeled example images, termed as "in-context examples", exploring content similarities between examples and the target. The resulting models can be generalized seamlessly to novel segmentation tasks, significantly reducing the labeling and training costs compared with conventional pipelines. However, in-context segmentation is more challenging than classic ones requiring the model to learn segmentation rules conditioned on a few samples. Unlike previous work with ad-hoc or non-end-to-end designs, we propose SEGIC, an end-to-end segment-in-context framework built upon a single vision foundation model (VFM). In particular, SEGIC leverages the emergent correspondence within VFM to capture dense relationships between target images and in-context samples. As such, information from in-context samples is then extracted into three types of instructions, i.e. geometric, visual, and meta instructions, serving as explicit conditions for the final mask prediction. SEGIC is a straightforward yet effective approach that yields state-of-the-art performance on one-shot segmentation benchmarks. Notably, SEGIC can be easily generalized to diverse tasks, including video object segmentation and open-vocabulary segmentation. Code will be available at https://github.com/MengLcool/SEGIC.

Tzofi Klinghoffer, Jonah Philion, Wenzheng Chen et al.

Autonomous vehicles (AV) require that neural networks used for perception be robust to different viewpoints if they are to be deployed across many types of vehicles without the repeated cost of data collection and labeling for each. AV companies typically focus on collecting data from diverse scenarios and locations, but not camera rig configurations, due to cost. As a result, only a small number of rig variations exist across most fleets. In this paper, we study how AV perception models are affected by changes in camera viewpoint and propose a way to scale them across vehicle types without repeated data collection and labeling. Using bird's eye view (BEV) segmentation as a motivating task, we find through extensive experiments that existing perception models are surprisingly sensitive to changes in camera viewpoint. When trained with data from one camera rig, small changes to pitch, yaw, depth, or height of the camera at inference time lead to large drops in performance. We introduce a technique for novel view synthesis and use it to transform collected data to the viewpoint of target rigs, allowing us to train BEV segmentation models for diverse target rigs without any additional data collection or labeling cost. To analyze the impact of viewpoint changes, we leverage synthetic data to mitigate other gaps (content, ISP, etc). Our approach is then trained on real data and evaluated on synthetic data, enabling evaluation on diverse target rigs. We release all data for use in future work. Our method is able to recover an average of 14.7% of the IoU that is otherwise lost when deploying to new rigs.

Jiayu Yang, Enze Xie, Miaomiao Liu et al.

Recent vision-only perception models for autonomous driving achieved promising results by encoding multi-view image features into Bird's-Eye-View (BEV) space. A critical step and the main bottleneck of these methods is transforming image features into the BEV coordinate frame. This paper focuses on leveraging geometry information, such as depth, to model such feature transformation. Existing works rely on non-parametric depth distribution modeling leading to significant memory consumption, or ignore the geometry information to address this problem. In contrast, we propose to use parametric depth distribution modeling for feature transformation. We first lift the 2D image features to the 3D space defined for the ego vehicle via a predicted parametric depth distribution for each pixel in each view. Then, we aggregate the 3D feature volume based on the 3D space occupancy derived from depth to the BEV frame. Finally, we use the transformed features for downstream tasks such as object detection and semantic segmentation. Existing semantic segmentation methods do also suffer from an hallucination problem as they do not take visibility information into account. This hallucination can be particularly problematic for subsequent modules such as control and planning. To mitigate the issue, our method provides depth uncertainty and reliable visibility-aware estimations. We further leverage our parametric depth modeling to present a novel visibility-aware evaluation metric that, when taken into account, can mitigate the hallucination problem. Extensive experiments on object detection and semantic segmentation on the nuScenes datasets demonstrate that our method outperforms existing methods on both tasks.

Hoang Chuong Nguyen, Renjie Wu, Jose M. Alvarez et al.

Generalizable 3D Gaussian Splatting has recently emerged as an efficient approach for novel-view synthesis, enabling feed-forward synthesis from only a few input views. However, existing pixel-wise feed-forward methods suffer from over-bright renderings when the number of input views varies during inference, as well as insufficient supervision for accurate Gaussian scale estimation, which leads to hole artifacts, particularly in high-resolution renderings. To address these issues, we identify that the over-brightness is caused by the varying number of overlapping Gaussians and propose a simple alpha normalization strategy to maintain brightness consistency across different number of input views. In addition, we introduce an auxiliary 3D sampling-based regularizer to improve Gaussian scale estimation, thereby mitigating hole artifacts in high-resolution rendering. Experiments on benchmark datasets demonstrate that our method significantly improves baseline models under varying input-view and high-resolution rendering settings.

Zhenxin Li, Shiyi Lan, Jose M. Alvarez et al.

Recently, the rise of query-based Transformer decoders is reshaping camera-based 3D object detection. These query-based decoders are surpassing the traditional dense BEV (Bird's Eye View)-based methods. However, we argue that dense BEV frameworks remain important due to their outstanding abilities in depth estimation and object localization, depicting 3D scenes accurately and comprehensively. This paper aims to address the drawbacks of the existing dense BEV-based 3D object detectors by introducing our proposed enhanced components, including a CRF-modulated depth estimation module enforcing object-level consistencies, a long-term temporal aggregation module with extended receptive fields, and a two-stage object decoder combining perspective techniques with CRF-modulated depth embedding. These enhancements lead to a "modernized" dense BEV framework dubbed BEVNeXt. On the nuScenes benchmark, BEVNeXt outperforms both BEV-based and query-based frameworks under various settings, achieving a state-of-the-art result of 64.2 NDS on the nuScenes test set. Code will be available at \url{https://github.com/woxihuanjiangguo/BEVNeXt}.

Maying Shen, Nadine Chang, Sifei Liu et al.

In recent years, the data collected for artificial intelligence has grown to an unmanageable amount. Particularly within industrial applications, such as autonomous vehicles, model training computation budgets are being exceeded while model performance is saturating -- and yet more data continues to pour in. To navigate the flood of data, we propose a framework to select the most semantically diverse and important dataset portion. Then, we further semantically enrich it by discovering meaningful new data from a massive unlabeled data pool. Importantly, we can provide explainability by leveraging foundation models to generate semantics for every data point. We quantitatively show that our Semantic Selection and Enrichment framework (SSE) can a) successfully maintain model performance with a smaller training dataset and b) improve model performance by enriching the smaller dataset without exceeding the original dataset size. Consequently, we demonstrate that semantic diversity is imperative for optimal data selection and model performance.

Jenny Schmalfuss, Nadine Chang, Vibashan VS et al.

Vision language models (VLMs) respond to user-crafted text prompts and visual inputs, and are applied to numerous real-world problems. VLMs integrate visual modalities with large language models (LLMs), which are well known to be prompt-sensitive. Hence, it is crucial to determine whether VLMs inherit this instability to varying prompts. We therefore investigate which prompt variations VLMs are most sensitive to and which VLMs are most agnostic to prompt variations. To this end, we introduce PARC (Prompt Analysis via Reliability and Calibration), a VLM prompt sensitivity analysis framework built on three pillars: (1) plausible prompt variations in both the language and vision domain, (2) a novel model reliability score with built-in guarantees, and (3) a calibration step that enables dataset- and prompt-spanning prompt variation analysis. Regarding prompt variations, PARC's evaluation shows that VLMs mirror LLM language prompt sensitivity in the vision domain, and most destructive variations change the expected answer. Regarding models, outstandingly robust VLMs among 22 evaluated models come from the InternVL2 family. We further find indications that prompt sensitivity is linked to training data. The code will be at https://github.com/NVlabs/PARC.

Hoang Chuong Nguyen, Wei Mao, Jose M. Alvarez et al.

Neural Radiance Fields (NeRF) has demonstrated its superior capability to represent 3D geometry but require accurately precomputed camera poses during training. To mitigate this requirement, existing methods jointly optimize camera poses and NeRF often relying on good pose initialisation or depth priors. However, these approaches struggle in challenging scenarios, such as large rotations, as they map each camera to a world coordinate system. We propose a novel method that eliminates prior dependencies by modeling continuous camera motions as time-dependent angular velocity and velocity. Relative motions between cameras are learned first via velocity integration, while camera poses can be obtained by aggregating such relative motions up to a world coordinate system defined at a single time step within the video. Specifically, accurate continuous camera movements are learned through a time-dependent NeRF, which captures local scene geometry and motion by training from neighboring frames for each time step. The learned motions enable fine-tuning the NeRF to represent the full scene geometry. Experiments on Co3D and Scannet show our approach achieves superior camera pose and depth estimation and comparable novel-view synthesis performance compared to state-of-the-art methods. Our code is available at https://github.com/HoangChuongNguyen/cope-nerf.

Zhiqi Li, Wenhai Wang, Enze Xie et al.

Panoptic segmentation involves a combination of joint semantic segmentation and instance segmentation, where image contents are divided into two types: things and stuff. We present Panoptic SegFormer, a general framework for panoptic segmentation with transformers. It contains three innovative components: an efficient deeply-supervised mask decoder, a query decoupling strategy, and an improved post-processing method. We also use Deformable DETR to efficiently process multi-scale features, which is a fast and efficient version of DETR. Specifically, we supervise the attention modules in the mask decoder in a layer-wise manner. This deep supervision strategy lets the attention modules quickly focus on meaningful semantic regions. It improves performance and reduces the number of required training epochs by half compared to Deformable DETR. Our query decoupling strategy decouples the responsibilities of the query set and avoids mutual interference between things and stuff. In addition, our post-processing strategy improves performance without additional costs by jointly considering classification and segmentation qualities to resolve conflicting mask overlaps. Our approach increases the accuracy 6.2\% PQ over the baseline DETR model. Panoptic SegFormer achieves state-of-the-art results on COCO test-dev with 56.2\% PQ. It also shows stronger zero-shot robustness over existing methods. The code is released at \url{https://github.com/zhiqi-li/Panoptic-SegFormer}.

Enze Xie, Wenhai Wang, Zhiding Yu et al.

We present SegFormer, a simple, efficient yet powerful semantic segmentation framework which unifies Transformers with lightweight multilayer perception (MLP) decoders. SegFormer has two appealing features: 1) SegFormer comprises a novel hierarchically structured Transformer encoder which outputs multiscale features. It does not need positional encoding, thereby avoiding the interpolation of positional codes which leads to decreased performance when the testing resolution differs from training. 2) SegFormer avoids complex decoders. The proposed MLP decoder aggregates information from different layers, and thus combining both local attention and global attention to render powerful representations. We show that this simple and lightweight design is the key to efficient segmentation on Transformers. We scale our approach up to obtain a series of models from SegFormer-B0 to SegFormer-B5, reaching significantly better performance and efficiency than previous counterparts. For example, SegFormer-B4 achieves 50.3% mIoU on ADE20K with 64M parameters, being 5x smaller and 2.2% better than the previous best method. Our best model, SegFormer-B5, achieves 84.0% mIoU on Cityscapes validation set and shows excellent zero-shot robustness on Cityscapes-C. Code will be released at: github.com/NVlabs/SegFormer.

Nadine Chang, Zhiding Yu, Yu-Xiong Wang et al.

Training on datasets with long-tailed distributions has been challenging for major recognition tasks such as classification and detection. To deal with this challenge, image resampling is typically introduced as a simple but effective approach. However, we observe that long-tailed detection differs from classification since multiple classes may be present in one image. As a result, image resampling alone is not enough to yield a sufficiently balanced distribution at the object level. We address object-level resampling by introducing an object-centric memory replay strategy based on dynamic, episodic memory banks. Our proposed strategy has two benefits: 1) convenient object-level resampling without significant extra computation, and 2) implicit feature-level augmentation from model updates. We show that image-level and object-level resamplings are both important, and thus unify them with a joint resampling strategy (RIO). Our method outperforms state-of-the-art long-tailed detection and segmentation methods on LVIS v0.5 across various backbones. Code is available at https://github.com/NVlabs/RIO.

Jiayu Yang, Jose M. Alvarez, Miaomiao Liu

Recent supervised multi-view depth estimation networks have achieved promising results. Similar to all supervised approaches, these networks require ground-truth data during training. However, collecting a large amount of multi-view depth data is very challenging. Here, we propose a self-supervised learning framework for multi-view stereo that exploit pseudo labels from the input data. We start by learning to estimate depth maps as initial pseudo labels under an unsupervised learning framework relying on image reconstruction loss as supervision. We then refine the initial pseudo labels using a carefully designed pipeline leveraging depth information inferred from higher resolution images and neighboring views. We use these high-quality pseudo labels as the supervision signal to train the network and improve, iteratively, its performance by self-training. Extensive experiments on the DTU dataset show that our proposed self-supervised learning framework outperforms existing unsupervised multi-view stereo networks by a large margin and performs on par compared to the supervised counterpart. Code is available at https://github.com/JiayuYANG/Self-supervised-CVP-MVSNet.

Ronak Kosti, Jose M. Alvarez, Adria Recasens et al.

In our everyday lives and social interactions we often try to perceive the emotional states of people. There has been a lot of research in providing machines with a similar capacity of recognizing emotions. From a computer vision perspective, most of the previous efforts have been focusing in analyzing the facial expressions and, in some cases, also the body pose. Some of these methods work remarkably well in specific settings. However, their performance is limited in natural, unconstrained environments. Psychological studies show that the scene context, in addition to facial expression and body pose, provides important information to our perception of people's emotions. However, the processing of the context for automatic emotion recognition has not been explored in depth, partly due to the lack of proper data. In this paper we present EMOTIC, a dataset of images of people in a diverse set of natural situations, annotated with their apparent emotion. The EMOTIC dataset combines two different types of emotion representation: (1) a set of 26 discrete categories, and (2) the continuous dimensions Valence, Arousal, and Dominance. We also present a detailed statistical and algorithmic analysis of the dataset along with annotators' agreement analysis. Using the EMOTIC dataset we train different CNN models for emotion recognition, combining the information of the bounding box containing the person with the contextual information extracted from the scene. Our results show how scene context provides important information to automatically recognize emotional states and motivate further research in this direction. Dataset and code is open-sourced and available at: https://github.com/rkosti/emotic and link for the peer-reviewed published article: https://ieeexplore.ieee.org/document/8713881

Hongxu Yin, Pavlo Molchanov, Zhizhong Li et al.

We introduce DeepInversion, a new method for synthesizing images from the image distribution used to train a deep neural network. We 'invert' a trained network (teacher) to synthesize class-conditional input images starting from random noise, without using any additional information about the training dataset. Keeping the teacher fixed, our method optimizes the input while regularizing the distribution of intermediate feature maps using information stored in the batch normalization layers of the teacher. Further, we improve the diversity of synthesized images using Adaptive DeepInversion, which maximizes the Jensen-Shannon divergence between the teacher and student network logits. The resulting synthesized images from networks trained on the CIFAR-10 and ImageNet datasets demonstrate high fidelity and degree of realism, and help enable a new breed of data-free applications - ones that do not require any real images or labeled data. We demonstrate the applicability of our proposed method to three tasks of immense practical importance -- (i) data-free network pruning, (ii) data-free knowledge transfer, and (iii) data-free continual learning. Code is available at https://github.com/NVlabs/DeepInversion

Jiayu Yang, Wei Mao, Jose M. Alvarez et al.

We propose a cost volume-based neural network for depth inference from multi-view images. We demonstrate that building a cost volume pyramid in a coarse-to-fine manner instead of constructing a cost volume at a fixed resolution leads to a compact, lightweight network and allows us inferring high resolution depth maps to achieve better reconstruction results. To this end, we first build a cost volume based on uniform sampling of fronto-parallel planes across the entire depth range at the coarsest resolution of an image. Then, given current depth estimate, we construct new cost volumes iteratively on the pixelwise depth residual to perform depth map refinement. While sharing similar insight with Point-MVSNet as predicting and refining depth iteratively, we show that working on cost volume pyramid can lead to a more compact, yet efficient network structure compared with the Point-MVSNet on 3D points. We further provide detailed analyses of the relation between (residual) depth sampling and image resolution, which serves as a principle for building compact cost volume pyramid. Experimental results on benchmark datasets show that our model can perform 6x faster and has similar performance as state-of-the-art methods. Code is available at https://github.com/JiayuYANG/CVP-MVSNet

Shuaiyi Huang, De-An Huang, Zhiding Yu et al.

Video instance segmentation (VIS) is a challenging vision task that aims to detect, segment, and track objects in videos. Conventional VIS methods rely on densely-annotated object masks which are expensive. We reduce the human annotations to only one point for each object in a video frame during training, and obtain high-quality mask predictions close to fully supervised models. Our proposed training method consists of a class-agnostic proposal generation module to provide rich negative samples and a spatio-temporal point-based matcher to match the object queries with the provided point annotations. Comprehensive experiments on three VIS benchmarks demonstrate competitive performance of the proposed framework, nearly matching fully supervised methods.

Nikita Durasov, Rafid Mahmood, Jiwoong Choi et al.

3D object detection is an essential task for computer vision applications in autonomous vehicles and robotics. However, models often struggle to quantify detection reliability, leading to poor performance on unfamiliar scenes. We introduce a framework for quantifying uncertainty in 3D object detection by leveraging an evidential learning loss on Bird's Eye View representations in the 3D detector. These uncertainty estimates require minimal computational overhead and are generalizable across different architectures. We demonstrate both the efficacy and importance of these uncertainty estimates on identifying out-of-distribution scenes, poorly localized objects, and missing (false negative) detections; our framework consistently improves over baselines by 10-20% on average. Finally, we integrate this suite of tasks into a system where a 3D object detector auto-labels driving scenes and our uncertainty estimates verify label correctness before the labels are used to train a second model. Here, our uncertainty-driven verification results in a 1% improvement in mAP and a 1-2% improvement in NDS.

Hoang Chuong Nguyen, Tianyu Wang, Jose M. Alvarez et al.

This paper focuses on self-supervised monocular depth estimation in dynamic scenes trained on monocular videos. Existing methods jointly estimate pixel-wise depth and motion, relying mainly on an image reconstruction loss. Dynamic regions1 remain a critical challenge for these methods due to the inherent ambiguity in depth and motion estimation, resulting in inaccurate depth estimation. This paper proposes a self-supervised training framework exploiting pseudo depth labels for dynamic regions from training data. The key contribution of our framework is to decouple depth estimation for static and dynamic regions of images in the training data. We start with an unsupervised depth estimation approach, which provides reliable depth estimates for static regions and motion cues for dynamic regions and allows us to extract moving object information at the instance level. In the next stage, we use an object network to estimate the depth of those moving objects assuming rigid motions. Then, we propose a new scale alignment module to address the scale ambiguity between estimated depths for static and dynamic regions. We can then use the depth labels generated to train an end-to-end depth estimation network and improve its performance. Extensive experiments on the Cityscapes and KITTI datasets show that our self-training strategy consistently outperforms existing self/unsupervised depth estimation methods.

Jingyu Song, Zhenxin Li, Shiyi Lan et al.

Benchmarking autonomous driving planners to align with human judgment remains a critical challenge, as state-of-the-art metrics like the Extended Predictive Driver Model Score (EPDMS) lack context awareness in nuanced scenarios. To address this, we introduce DriveCritic, a novel framework featuring two key contributions: the DriveCritic dataset, a curated collection of challenging scenarios where context is critical for correct judgment and annotated with pairwise human preferences, and the DriveCritic model, a Vision-Language Model (VLM) based evaluator. Fine-tuned using a two-stage supervised and reinforcement learning pipeline, the DriveCritic model learns to adjudicate between trajectory pairs by integrating visual and symbolic context. Experiments show DriveCritic significantly outperforms existing metrics and baselines in matching human preferences and demonstrates strong context awareness. Overall, our work provides a more reliable, human-aligned foundation to evaluating autonomous driving systems.

Shan Wang, Maying Shen, Nadine Chang et al.

Multimodal large language models achieve strong performance across diverse tasks but remain prone to hallucinations, where outputs are not grounded in visual inputs. This issue can be attributed to two main biases: text-visual bias, the overreliance on prompts and prior outputs, and co-occurrence bias, spurious correlations between frequently paired objects. We propose Gradient-based Influence-Aware Constrained Decoding (GACD), an inference-based method, that addresses both biases without auxiliary models, and is readily applicable to existing models without finetuning. The core of our approach is bias estimation, which uses first-order Taylor gradients to understand the contribution of individual tokens-visual features and text tokens-to the current output. Based on this analysis, GACD mitigates hallucinations through two components: (1) suppressing spurious visual features correlated with the output objects, and (2) rebalancing cross-modal contributions by strengthening visual features relative to text. Experiments across multiple benchmarks demonstrate that GACD effectively reduces hallucinations and improves the visual grounding of MLLM outputs.

Xinnan Du, William Zhang, Jose M. Alvarez

Deep learning perception models require a massive amount of labeled training data to achieve good performance. While unlabeled data is easy to acquire, the cost of labeling is prohibitive and could create a tremendous burden on companies or individuals. Recently, self-supervision has emerged as an alternative to leveraging unlabeled data. In this paper, we propose a new light-weight self-supervised learning framework that could boost supervised learning performance with minimum additional computation cost. Here, we introduce a simple and flexible multi-task co-training framework that integrates a self-supervised task into any supervised task. Our approach exploits pretext tasks to incur minimum compute and parameter overheads and minimal disruption to existing training pipelines. We demonstrate the effectiveness of our framework by using two self-supervised tasks, object detection and panoptic segmentation, on different perception models. Our results show that both self-supervised tasks can improve the accuracy of the supervised task and, at the same time, demonstrates strong domain adaption capability when used with additional unlabeled data.

Maying Shen, Pavlo Molchanov, Hongxu Yin et al.

Pruning enables appealing reductions in network memory footprint and time complexity. Conventional post-training pruning techniques lean towards efficient inference while overlooking the heavy computation for training. Recent exploration of pre-training pruning at initialization hints on training cost reduction via pruning, but suffers noticeable performance degradation. We attempt to combine the benefits of both directions and propose a policy that prunes as early as possible during training without hurting performance. Instead of pruning at initialization, our method exploits initial dense training for few epochs to quickly guide the architecture, while constantly evaluating dominant sub-networks via neuron importance ranking. This unveils dominant sub-networks whose structures turn stable, allowing conventional pruning to be pushed earlier into the training. To do this early, we further introduce an Early Pruning Indicator (EPI) that relies on sub-network architectural similarity and quickly triggers pruning when the sub-network's architecture stabilizes. Through extensive experiments on ImageNet, we show that EPI empowers a quick tracking of early training epochs suitable for pruning, offering same efficacy as an otherwise ``oracle'' grid-search that scans through epochs and requires orders of magnitude more compute. Our method yields $1.4\%$ top-1 accuracy boost over state-of-the-art pruning counterparts, cuts down training cost on GPU by $2.4\times$, hence offers a new efficiency-accuracy boundary for network pruning during training.

Maying Shen, Hongxu Yin, Pavlo Molchanov et al.

Structural pruning can simplify network architecture and improve inference speed. We propose Hardware-Aware Latency Pruning (HALP) that formulates structural pruning as a global resource allocation optimization problem, aiming at maximizing the accuracy while constraining latency under a predefined budget. For filter importance ranking, HALP leverages latency lookup table to track latency reduction potential and global saliency score to gauge accuracy drop. Both metrics can be evaluated very efficiently during pruning, allowing us to reformulate global structural pruning under a reward maximization problem given target constraint. This makes the problem solvable via our augmented knapsack solver, enabling HALP to surpass prior work in pruning efficacy and accuracy-efficiency trade-off. We examine HALP on both classification and detection tasks, over varying networks, on ImageNet and VOC datasets. In particular, for ResNet-50/-101 pruning on ImageNet, HALP improves network throughput by $1.60\times$/$1.90\times$ with $+0.3\%$/$-0.2\%$ top-1 accuracy changes, respectively. For SSD pruning on VOC, HALP improves throughput by $1.94\times$ with only a $0.56$ mAP drop. HALP consistently outperforms prior art, sometimes by large margins.

Hongxu Yin, Arun Mallya, Arash Vahdat et al.

Training deep neural networks requires gradient estimation from data batches to update parameters. Gradients per parameter are averaged over a set of data and this has been presumed to be safe for privacy-preserving training in joint, collaborative, and federated learning applications. Prior work only showed the possibility of recovering input data given gradients under very restrictive conditions - a single input point, or a network with no non-linearities, or a small 32x32 px input batch. Therefore, averaging gradients over larger batches was thought to be safe. In this work, we introduce GradInversion, using which input images from a larger batch (8 - 48 images) can also be recovered for large networks such as ResNets (50 layers), on complex datasets such as ImageNet (1000 classes, 224x224 px). We formulate an optimization task that converts random noise into natural images, matching gradients while regularizing image fidelity. We also propose an algorithm for target class label recovery given gradients. We further propose a group consistency regularization framework, where multiple agents starting from different random seeds work together to find an enhanced reconstruction of original data batch. We show that gradients encode a surprisingly large amount of information, such that all the individual images can be recovered with high fidelity via GradInversion, even for complex datasets, deep networks, and large batch sizes.

Wuyang Chen, Zhiding Yu, Shalini De Mello et al.

Training on synthetic data can be beneficial for label or data-scarce scenarios. However, synthetically trained models often suffer from poor generalization in real domains due to domain gaps. In this work, we make a key observation that the diversity of the learned feature embeddings plays an important role in the generalization performance. To this end, we propose contrastive synthetic-to-real generalization (CSG), a novel framework that leverages the pre-trained ImageNet knowledge to prevent overfitting to the synthetic domain, while promoting the diversity of feature embeddings as an inductive bias to improve generalization. In addition, we enhance the proposed CSG framework with attentional pooling (A-pool) to let the model focus on semantically important regions and further improve its generalization. We demonstrate the effectiveness of CSG on various synthetic training tasks, exhibiting state-of-the-art performance on zero-shot domain generalization.

Jiwoong Choi, Ismail Elezi, Hyuk-Jae Lee et al.

Active learning aims to reduce labeling costs by selecting only the most informative samples on a dataset. Few existing works have addressed active learning for object detection. Most of these methods are based on multiple models or are straightforward extensions of classification methods, hence estimate an image's informativeness using only the classification head. In this paper, we propose a novel deep active learning approach for object detection. Our approach relies on mixture density networks that estimate a probabilistic distribution for each localization and classification head's output. We explicitly estimate the aleatoric and epistemic uncertainty in a single forward pass of a single model. Our method uses a scoring function that aggregates these two types of uncertainties for both heads to obtain every image's informativeness score. We demonstrate the efficacy of our approach in PASCAL VOC and MS-COCO datasets. Our approach outperforms single-model based methods and performs on par with multi-model based methods at a fraction of the computing cost.

Michael Zhang, Karan Sapra, Sanja Fidler et al.

While federated learning traditionally aims to train a single global model across decentralized local datasets, one model may not always be ideal for all participating clients. Here we propose an alternative, where each client only federates with other relevant clients to obtain a stronger model per client-specific objectives. To achieve this personalization, rather than computing a single model average with constant weights for the entire federation as in traditional FL, we efficiently calculate optimal weighted model combinations for each client, based on figuring out how much a client can benefit from another's model. We do not assume knowledge of any underlying data distributions or client similarities, and allow each client to optimize for arbitrary target distributions of interest, enabling greater flexibility for personalization. We evaluate and characterize our method on a variety of federated settings, datasets, and degrees of local data heterogeneity. Our method outperforms existing alternatives, while also enabling new features for personalized FL such as transfer outside of local data distributions.

Elmar Haussmann, Michele Fenzi, Kashyap Chitta et al.

Deep Neural Networks trained in a fully supervised fashion are the dominant technology in perception-based autonomous driving systems. While collecting large amounts of unlabeled data is already a major undertaking, only a subset of it can be labeled by humans due to the effort needed for high-quality annotation. Therefore, finding the right data to label has become a key challenge. Active learning is a powerful technique to improve data efficiency for supervised learning methods, as it aims at selecting the smallest possible training set to reach a required performance. We have built a scalable production system for active learning in the domain of autonomous driving. In this paper, we describe the resulting high-level design, sketch some of the challenges and their solutions, present our current results at scale, and briefly describe the open problems and future directions.

Shuang Gao, Xin Liu, Lung-Sheng Chien et al.

Improving weight sparsity is a common strategy for producing light-weight deep neural networks. However, pruning models with residual learning is more challenging. In this paper, we introduce Variance-Aware Cross-Layer (VACL), a novel approach to address this problem. VACL consists of two parts, a Cross-Layer grouping and a Variance Aware regularization. In Cross-Layer grouping the $i^{th}$ filters of layers connected by skip-connections are grouped into one regularization group. Then, the Variance-Aware regularization term takes into account both the first and second-order statistics of the connected layers to constrain the variance within a group. Our approach can effectively improve the structural sparsity of residual models. For CIFAR10, the proposed method reduces a ResNet model by up to 79.5% with no accuracy drop and reduces a ResNeXt model by up to 82% with less than 1% accuracy drop. For ImageNet, it yields a pruned ratio of up to 63.3% with less than 1% top-5 accuracy drop. Our experimental results show that the proposed approach significantly outperforms other state-of-the-art methods in terms of overall model size and accuracy.

Kashyap Chitta, Jose M. Alvarez, Elmar Haussmann et al.

Deep Neural Networks (DNNs) often rely on very large datasets for training. Given the large size of such datasets, it is conceivable that they contain certain samples that either do not contribute or negatively impact the DNN's optimization. Modifying the training distribution in a way that excludes such samples could provide an effective solution to both improve performance and reduce training time. In this paper, we propose to scale up ensemble Active Learning (AL) methods to perform acquisition at a large scale (10k to 500k samples at a time). We do this with ensembles of hundreds of models, obtained at a minimal computational cost by reusing intermediate training checkpoints. This allows us to automatically and efficiently perform a training data subset search for large labeled datasets. We observe that our approach obtains favorable subsets of training data, which can be used to train more accurate DNNs than training with the entire dataset. We perform an extensive experimental study of this phenomenon on three image classification benchmarks (CIFAR-10, CIFAR-100 and ImageNet), as well as an internal object detection benchmark for prototyping perception models for autonomous driving. Unlike existing studies, our experiments on object detection are at the scale required for production-ready autonomous driving systems. We provide insights on the impact of different initialization schemes, acquisition functions and ensemble configurations at this scale. Our results provide strong empirical evidence that optimizing the training data distribution can provide significant benefits on large scale vision tasks.

Jiaming Zeng, Adam Lesnikowski, Jose M. Alvarez

One of the main challenges of deep learning tools is their inability to capture model uncertainty. While Bayesian deep learning can be used to tackle the problem, Bayesian neural networks often require more time and computational power to train than deterministic networks. Our work explores whether fully Bayesian networks are needed to successfully capture model uncertainty. We vary the number and position of Bayesian layers in a network and compare their performance on active learning with the MNIST dataset. We found that we can fully capture the model uncertainty by using only a few Bayesian layers near the output of the network, combining the advantages of deterministic and Bayesian networks.

Shuxuan Guo, Jose M. Alvarez, Mathieu Salzmann

We introduce an approach to training a given compact network. To this end, we leverage over-parameterization, which typically improves both neural network optimization and generalization. Specifically, we propose to expand each linear layer of the compact network into multiple consecutive linear layers, without adding any nonlinearity. As such, the resulting expanded network, or ExpandNet, can be contracted back to the compact one algebraically at inference. In particular, we introduce two convolutional expansion strategies and demonstrate their benefits on several tasks, including image classification, object detection, and semantic segmentation. As evidenced by our experiments, our approach outperforms both training the compact network from scratch and performing knowledge distillation from a teacher. Furthermore, our linear over-parameterization empirically reduces gradient confusion during training and improves the network generalization.

Kashyap Chitta, Jose M. Alvarez, Adam Lesnikowski

Annotating the right data for training deep neural networks is an important challenge. Active learning using uncertainty estimates from Bayesian Neural Networks (BNNs) could provide an effective solution to this. Despite being theoretically principled, BNNs require approximations to be applied to large-scale problems, where both performance and uncertainty estimation are crucial. In this paper, we introduce Deep Probabilistic Ensembles (DPEs), a scalable technique that uses a regularized ensemble to approximate a deep BNN. We conduct a series of large-scale visual active learning experiments to evaluate DPEs on classification with the CIFAR-10, CIFAR-100 and ImageNet datasets, and semantic segmentation with the BDD100k dataset. Our models require significantly less training data to achieve competitive performances, and steadily improve upon strong active learning baselines as the annotation budget is increased.

Kashyap Chitta, Jose M. Alvarez, Adam Lesnikowski

In this paper, we introduce Deep Probabilistic Ensembles (DPEs), a scalable technique that uses a regularized ensemble to approximate a deep Bayesian Neural Network (BNN). We do so by incorporating a KL divergence penalty term into the training objective of an ensemble, derived from the evidence lower bound used in variational inference. We evaluate the uncertainty estimates obtained from our models for active learning on visual classification. Our approach steadily improves upon active learning baselines as the annotation budget is increased.

Fatemeh Sadat Saleh, Mohammad Sadegh Aliakbarian, Mathieu Salzmann et al.

Training a deep network to perform semantic segmentation requires large amounts of labeled data. To alleviate the manual effort of annotating real images, researchers have investigated the use of synthetic data, which can be labeled automatically. Unfortunately, a network trained on synthetic data performs relatively poorly on real images. While this can be addressed by domain adaptation, existing methods all require having access to real images during training. In this paper, we introduce a drastically different way to handle synthetic images that does not require seeing any real images at training time. Our approach builds on the observation that foreground and background classes are not affected in the same manner by the domain shift, and thus should be treated differently. In particular, the former should be handled in a detection-based manner to better account for the fact that, while their texture in synthetic images is not photo-realistic, their shape looks natural. Our experiments evidence the effectiveness of our approach on Cityscapes and CamVid with models trained on synthetic data only.

Jose M. Alvarez, Mathieu Salzmann

In recent years, great progress has been made in a variety of application domains thanks to the development of increasingly deeper neural networks. Unfortunately, the huge number of units of these networks makes them expensive both computationally and memory-wise. To overcome this, exploiting the fact that deep networks are over-parametrized, several compression strategies have been proposed. These methods, however, typically start from a network that has been trained in a standard manner, without considering such a future compression. In this paper, we propose to explicitly account for compression in the training process. To this end, we introduce a regularizer that encourages the parameter matrix of each layer to have low rank during training. We show that accounting for compression during training allows us to learn much more compact, yet at least as effective, models than state-of-the-art compression techniques.

Marc Masana, Joost van de Weijer, Luis Herranz et al.

Deep Neural Networks trained on large datasets can be easily transferred to new domains with far fewer labeled examples by a process called fine-tuning. This has the advantage that representations learned in the large source domain can be exploited on smaller target domains. However, networks designed to be optimal for the source task are often prohibitively large for the target task. In this work we address the compression of networks after domain transfer. We focus on compression algorithms based on low-rank matrix decomposition. Existing methods base compression solely on learned network weights and ignore the statistics of network activations. We show that domain transfer leads to large shifts in network activations and that it is desirable to take this into account when compressing. We demonstrate that considering activation statistics when compressing weights leads to a rank-constrained regression problem with a closed-form solution. Because our method takes into account the target domain, it can more optimally remove the redundancy in the weights. Experiments show that our Domain Adaptive Low Rank (DALR) method significantly outperforms existing low-rank compression techniques. With our approach, the fc6 layer of VGG19 can be compressed more than 4x more than using truncated SVD alone -- with only a minor or no loss in accuracy. When applied to domain-transferred networks it allows for compression down to only 5-20% of the original number of parameters with only a minor drop in performance.

Fatemeh Sadat Saleh, Mohammad Sadegh Aliakbarian, Mathieu Salzmann et al.

Pixel-level annotations are expensive and time-consuming to obtain. Hence, weak supervision using only image tags could have a significant impact in semantic segmentation. Recent years have seen great progress in weakly-supervised semantic segmentation, whether from a single image or from videos. However, most existing methods are designed to handle a single background class. In practical applications, such as autonomous navigation, it is often crucial to reason about multiple background classes. In this paper, we introduce an approach to doing so by making use of classifier heatmaps. We then develop a two-stream deep architecture that jointly leverages appearance and motion, and design a loss based on our heatmaps to train it. Our experiments demonstrate the benefits of our classifier heatmaps and of our two-stream architecture on challenging urban scene datasets and on the YouTube-Objects benchmark, where we obtain state-of-the-art results.

Albert Jimenez, Jose M. Alvarez, Xavier Giro-i-Nieto

Image retrieval in realistic scenarios targets large dynamic datasets of unlabeled images. In these cases, training or fine-tuning a model every time new images are added to the database is neither efficient nor scalable. Convolutional neural networks trained for image classification over large datasets have been proven effective feature extractors for image retrieval. The most successful approaches are based on encoding the activations of convolutional layers, as they convey the image spatial information. In this paper, we go beyond this spatial information and propose a local-aware encoding of convolutional features based on semantic information predicted in the target image. To this end, we obtain the most discriminative regions of an image using Class Activation Maps (CAMs). CAMs are based on the knowledge contained in the network and therefore, our approach, has the additional advantage of not requiring external information. In addition, we use CAMs to generate object proposals during an unsupervised re-ranking stage after a first fast search. Our experiments on two public available datasets for instance retrieval, Oxford5k and Paris6k, demonstrate the competitiveness of our approach outperforming the current state-of-the-art when using off-the-shelf models trained on ImageNet. The source code and model used in this paper are publicly available at http://imatge-upc.github.io/retrieval-2017-cam/.

Fatemeh Sadat Saleh, Mohammad Sadegh Aliakbarian, Mathieu Salzmann et al.

Pixel-level annotations are expensive and time consuming to obtain. Hence, weak supervision using only image tags could have a significant impact in semantic segmentation. Recently, CNN-based methods have proposed to fine-tune pre-trained networks using image tags. Without additional information, this leads to poor localization accuracy. This problem, however, was alleviated by making use of objectness priors to generate foreground/background masks. Unfortunately these priors either require pixel-level annotations/bounding boxes, or still yield inaccurate object boundaries. Here, we propose a novel method to extract accurate masks from networks pre-trained for the task of object recognition, thus forgoing external objectness modules. We first show how foreground/background masks can be obtained from the activations of higher-level convolutional layers of a network. We then show how to obtain multi-class masks by the fusion of foreground/background ones with information extracted from a weakly-supervised localization network. Our experiments evidence that exploiting these masks in conjunction with a weakly-supervised training loss yields state-of-the-art tag-based weakly-supervised semantic segmentation results.

Guim Perarnau, Joost van de Weijer, Bogdan Raducanu et al.

Generative Adversarial Networks (GANs) have recently demonstrated to successfully approximate complex data distributions. A relevant extension of this model is conditional GANs (cGANs), where the introduction of external information allows to determine specific representations of the generated images. In this work, we evaluate encoders to inverse the mapping of a cGAN, i.e., mapping a real image into a latent space and a conditional representation. This allows, for example, to reconstruct and modify real images of faces conditioning on arbitrary attributes. Additionally, we evaluate the design of cGANs. The combination of an encoder with a cGAN, which we call Invertible cGAN (IcGAN), enables to re-generate real images with deterministic complex modifications.

Jose M Alvarez, Mathieu Salzmann

Nowadays, the number of layers and of neurons in each layer of a deep network are typically set manually. While very deep and wide networks have proven effective in general, they come at a high memory and computation cost, thus making them impractical for constrained platforms. These networks, however, are known to have many redundant parameters, and could thus, in principle, be replaced by more compact architectures. In this paper, we introduce an approach to automatically determining the number of neurons in each layer of a deep network during learning. To this end, we propose to make use of structured sparsity during learning. More precisely, we use a group sparsity regularizer on the parameters of the network, where each group is defined to act on a single neuron. Starting from an overcomplete network, we show that our approach can reduce the number of parameters by up to 80\% while retaining or even improving the network accuracy.

Fatemehsadat Saleh, Mohammad Sadegh Ali Akbarian, Mathieu Salzmann et al.

Pixel-level annotations are expensive and time consuming to obtain. Hence, weak supervision using only image tags could have a significant impact in semantic segmentation. Recently, CNN-based methods have proposed to fine-tune pre-trained networks using image tags. Without additional information, this leads to poor localization accuracy. This problem, however, was alleviated by making use of objectness priors to generate foreground/background masks. Unfortunately these priors either require training pixel-level annotations/bounding boxes, or still yield inaccurate object boundaries. Here, we propose a novel method to extract markedly more accurate masks from the pre-trained network itself, forgoing external objectness modules. This is accomplished using the activations of the higher-level convolutional layers, smoothed by a dense CRF. We demonstrate that our method, based on these masks and a weakly-supervised loss, outperforms the state-of-the-art tag-based weakly-supervised semantic segmentation techniques. Furthermore, we introduce a new form of inexpensive weak supervision yielding an additional accuracy boost.

Jose Alvarez, Lars Petersson

Deep learning and convolutional neural networks (ConvNets) have been successfully applied to most relevant tasks in the computer vision community. However, these networks are computationally demanding and not suitable for embedded devices where memory and time consumption are relevant. In this paper, we propose DecomposeMe, a simple but effective technique to learn features using 1D convolutions. The proposed architecture enables both simplicity and filter sharing leading to increased learning capacity. A comprehensive set of large-scale experiments on ImageNet and Places2 demonstrates the ability of our method to improve performance while significantly reducing the number of parameters required. Notably, on Places2, we obtain an improvement in relative top-1 classification accuracy of 7.7\% with an architecture that requires 92% fewer parameters compared to VGG-B. The proposed network is also demonstrated to generalize to other tasks by converting existing networks.

Gucan Long, Laurent Kneip, Jose M. Alvarez et al.

This work presents an unsupervised learning based approach to the ubiquitous computer vision problem of image matching. We start from the insight that the problem of frame-interpolation implicitly solves for inter-frame correspondences. This permits the application of analysis-by-synthesis: we firstly train and apply a Convolutional Neural Network for frame-interpolation, then obtain correspondences by inverting the learned CNN. The key benefit behind this strategy is that the CNN for frame-interpolation can be trained in an unsupervised manner by exploiting the temporal coherency that is naturally contained in real-world video sequences. The present model therefore learns image matching by simply watching videos. Besides a promise to be more generally applicable, the presented approach achieves surprising performance comparable to traditional empirically designed methods.

Jose M. Alvarez, Theo Gevers, Antonio M. Lopez

Detecting traversable road areas ahead a moving vehicle is a key process for modern autonomous driving systems. A common approach to road detection consists of exploiting color features to classify pixels as road or background. These algorithms reduce the effect of lighting variations and weather conditions by exploiting the discriminant/invariant properties of different color representations. Furthermore, the lack of labeled datasets has motivated the development of algorithms performing on single images based on the assumption that the bottom part of the image belongs to the road surface. In this paper, we first introduce a dataset of road images taken at different times and in different scenarios using an onboard camera. Then, we devise a simple online algorithm and conduct an exhaustive evaluation of different classifiers and the effect of using different color representation to characterize pixels.