One Step at a Time: Long-Horizon Vision-and-Language Navigation with Milestones

We study the problem of developing autonomous agents that can follow human instructions to infer and perform a sequence of actions to complete the underlying task. Significant progress has been made in recent years, especially for tasks with short horizons. However, when it comes to long-horizon tasks with extended sequences of actions, an agent can easily ignore some instructions or get stuck in the middle of the long instructions and eventually fail the task. To address this challenge, we propose a model-agnostic milestone-based task tracker (M-TRACK) to guide the agent and monitor its progress. Specifically, we propose a milestone builder that tags the instructions with navigation and interaction milestones which the agent needs to complete step by step, and a milestone checker that systemically checks the agent's progress in its current milestone and determines when to proceed to the next. On the challenging ALFRED dataset, our M-TRACK leads to a notable 45% and 70% relative improvement in unseen success rate over two competitive base models.

Two-Stage Mesh Deep Learning for Automated Tooth Segmentation and Landmark Localization on 3D Intraoral Scans

Accurately segmenting teeth and identifying the corresponding anatomical landmarks on dental mesh models are essential in computer-aided orthodontic treatment. Manually performing these two tasks is time-consuming, tedious, and, more importantly, highly dependent on orthodontists' experiences due to the abnormality and large-scale variance of patients' teeth. Some machine learning-based methods have been designed and applied in the orthodontic field to automatically segment dental meshes (e.g., intraoral scans). In contrast, the number of studies on tooth landmark localization is still limited. This paper proposes a two-stage framework based on mesh deep learning (called TS-MDL) for joint tooth labeling and landmark identification on raw intraoral scans. Our TS-MDL first adopts an end-to-end \emph{i}MeshSegNet method (i.e., a variant of the existing MeshSegNet with both improved accuracy and efficiency) to label each tooth on the downsampled scan. Guided by the segmentation outputs, our TS-MDL further selects each tooth's region of interest (ROI) on the original mesh to construct a light-weight variant of the pioneering PointNet (i.e., PointNet-Reg) for regressing the corresponding landmark heatmaps. Our TS-MDL was evaluated on a real-clinical dataset, showing promising segmentation and localization performance. Specifically, \emph{i}MeshSegNet in the first stage of TS-MDL reached an averaged Dice similarity coefficient (DSC) at $0.953\pm0.076$, significantly outperforming the original MeshSegNet. In the second stage, PointNet-Reg achieved a mean absolute error (MAE) of $0.623\pm0.718 \, mm$ in distances between the prediction and ground truth for $44$ landmarks, which is superior compared with other networks for landmark detection. All these results suggest the potential usage of our TS-MDL in clinical practices.

Sequential Joint Shape and Pose Estimation of Vehicles with Application to Automatic Amodal Segmentation Labeling

Shape and pose estimation is a critical perception problem for a self-driving car to fully understand its surrounding environment. One fundamental challenge in solving this problem is the incomplete sensor signal (e.g., LiDAR scans), especially for faraway or occluded objects. In this paper, we propose a novel algorithm to address this challenge, we explicitly leverage the sensor signal captured over consecutive time: the consecutive signals can provide more information of an object, including different viewpoints and its motion. By encoding the consecutive signal via a recurrent neural network, our algorithm not only improves shape and pose estimation, but also produces a labeling tool that can benefit other tasks in autonomous driving research. Specifically, building upon our algorithm, we propose a novel pipeline to automatically annotate high-quality labels for amodal segmentation on images, which are hard and laborious to annotate manually. Our code and data will be made publicly available.

Discovering the Unknown Knowns: Turning Implicit Knowledge in the Dataset into Explicit Training Examples for Visual Question Answering

Visual question answering (VQA) is challenging not only because the model has to handle multi-modal information, but also because it is just so hard to collect sufficient training examples -- there are too many questions one can ask about an image. As a result, a VQA model trained solely on human-annotated examples could easily over-fit specific question styles or image contents that are being asked, leaving the model largely ignorant about the sheer diversity of questions. Existing methods address this issue primarily by introducing an auxiliary task such as visual grounding, cycle consistency, or debiasing. In this paper, we take a drastically different approach. We found that many of the "unknowns" to the learned VQA model are indeed "known" in the dataset implicitly. For instance, questions asking about the same object in different images are likely paraphrases; the number of detected or annotated objects in an image already provides the answer to the "how many" question, even if the question has not been annotated for that image. Building upon these insights, we present a simple data augmentation pipeline SimpleAug to turn this "known" knowledge into training examples for VQA. We show that these augmented examples can notably improve the learned VQA models' performance, not only on the VQA-CP dataset with language prior shifts but also on the VQA v2 dataset without such shifts. Our method further opens up the door to leverage weakly-labeled or unlabeled images in a principled way to enhance VQA models. Our code and data are publicly available at https://github.com/heendung/simpleAUG.

On Model Calibration for Long-Tailed Object Detection and Instance Segmentation

Vanilla models for object detection and instance segmentation suffer from the heavy bias toward detecting frequent objects in the long-tailed setting. Existing methods address this issue mostly during training, e.g., by re-sampling or re-weighting. In this paper, we investigate a largely overlooked approach -- post-processing calibration of confidence scores. We propose NorCal, Normalized Calibration for long-tailed object detection and instance segmentation, a simple and straightforward recipe that reweighs the predicted scores of each class by its training sample size. We show that separately handling the background class and normalizing the scores over classes for each proposal are keys to achieving superior performance. On the LVIS dataset, NorCal can effectively improve nearly all the baseline models not only on rare classes but also on common and frequent classes. Finally, we conduct extensive analysis and ablation studies to offer insights into various modeling choices and mechanisms of our approach.

On Bridging Generic and Personalized Federated Learning

Federated learning is promising for its ability to collaboratively train models with multiple clients without accessing their data, but vulnerable when clients' data distributions diverge from each other. This divergence further leads to a dilemma: "Should we prioritize the learned model's generic performance (for future use at the server) or its personalized performance (for each client)?" These two, seemingly competing goals have divided the community to focus on one or the other, yet in this paper we show that it is possible to approach both at the same time. Concretely, we propose a novel federated learning framework that explicitly decouples a model's dual duties with two prediction tasks. On the one hand, we introduce a family of losses that are robust to non-identical class distributions, enabling clients to train a generic predictor with a consistent objective across them. On the other hand, we formulate the personalized predictor as a lightweight adaptive module that is learned to minimize each client's empirical risk on top of the generic predictor. With this two-loss, two-predictor framework which we name Federated Robust Decoupling Fed-RoD, the learned model can simultaneously achieve state-of-the-art generic and personalized performance, essentially bridging the two tasks.

Few-Shot Learning with a Strong Teacher

Few-shot learning (FSL) aims to train a strong classifier using limited labeled examples. Many existing works take the meta-learning approach, sampling few-shot tasks in turn and optimizing the few-shot learner's performance on classifying the query examples. In this paper, we point out two potential weaknesses of this approach. First, the sampled query examples may not provide sufficient supervision for the few-shot learner. Second, the effectiveness of meta-learning diminishes sharply with increasing shots (i.e., the number of training examples per class). To resolve these issues, we propose a novel objective to directly train the few-shot learner to perform like a strong classifier. Concretely, we associate each sampled few-shot task with a strong classifier, which is learned with ample labeled examples. The strong classifier has a better generalization ability and we use it to supervise the few-shot learner. We present an efficient way to construct the strong classifier, making our proposed objective an easily plug-and-play term to existing meta-learning based FSL methods. We validate our approach in combinations with many representative meta-learning methods. On several benchmark datasets including miniImageNet and tiredImageNet, our approach leads to a notable improvement across a variety of tasks. More importantly, with our approach, meta-learning based FSL methods can consistently outperform non-meta-learning based ones, even in a many-shot setting, greatly strengthening their applicability.

How to Train Your MAML to Excel in Few-Shot Classification

Model-agnostic meta-learning (MAML) is arguably the most popular meta-learning algorithm nowadays, given its flexibility to incorporate various model architectures and to be applied to different problems. Nevertheless, its performance on few-shot classification is far behind many recent algorithms dedicated to the problem. In this paper, we point out several key facets of how to train MAML to excel in few-shot classification. First, we find that a large number of gradient steps are needed for the inner loop update, which contradicts the common usage of MAML for few-shot classification. Second, we find that MAML is sensitive to the permutation of class assignments in meta-testing: for a few-shot task of $N$ classes, there are exponentially many ways to assign the learned initialization of the $N$-way classifier to the $N$ classes, leading to an unavoidably huge variance. Third, we investigate several ways for permutation invariance and find that learning a shared classifier initialization for all the classes performs the best. On benchmark datasets such as MiniImageNet and TieredImageNet, our approach, which we name UNICORN-MAML, performs on a par with or even outperforms state-of-the-art algorithms, while keeping the simplicity of MAML without adding any extra sub-networks.

Revisiting Document Representations for Large-Scale Zero-Shot Learning

Zero-shot learning aims to recognize unseen objects using their semantic representations. Most existing works use visual attributes labeled by humans, not suitable for large-scale applications. In this paper, we revisit the use of documents as semantic representations. We argue that documents like Wikipedia pages contain rich visual information, which however can easily be buried by the vast amount of non-visual sentences. To address this issue, we propose a semi-automatic mechanism for visual sentence extraction that leverages the document section headers and the clustering structure of visual sentences. The extracted visual sentences, after a novel weighting scheme to distinguish similar classes, essentially form semantic representations like visual attributes but need much less human effort. On the ImageNet dataset with over 10,000 unseen classes, our representations lead to a 64% relative improvement against the commonly used ones.

Procrustean Training for Imbalanced Deep Learning

Neural networks trained with class-imbalanced data are known to perform poorly on minor classes of scarce training data. Several recent works attribute this to over-fitting to minor classes. In this paper, we provide a novel explanation of this issue. We found that a neural network tends to first under-fit the minor classes by classifying most of their data into the major classes in early training epochs. To correct these wrong predictions, the neural network then must focus on pushing features of minor class data across the decision boundaries between major and minor classes, leading to much larger gradients for features of minor classes. We argue that such an under-fitting phase over-emphasizes the competition between major and minor classes, hinders the neural network from learning the discriminative knowledge that can be generalized to test data, and eventually results in over-fitting. To address this issue, we propose a novel learning strategy to equalize the training progress across classes. We mix features of the major class data with those of other data in a mini-batch, intentionally weakening their features to prevent a neural network from fitting them first. We show that this strategy can largely balance the training accuracy and feature gradients across classes, effectively mitigating the under-fitting then over-fitting problem for minor class data. On several benchmark datasets, our approach achieves the state-of-the-art accuracy, especially for the challenging step-imbalanced cases.