Few-shot, fine-grained classification requires a model to learn subtle, fine-grained distinctions between different classes (e.g., birds) based on a few images alone. This requires a remarkable degree of invariance to pose, articulation and background. A solution is to use pose-normalized representations: first localize semantic parts in each image, and then describe images by characterizing the appearance of each part. While such representations are out of favor for fully supervised classification, we show that they are extremely effective for few-shot fine-grained classification. With a minimal increase in model capacity, pose normalization improves accuracy between 10 and 20 percentage points for shallow and deep architectures, generalizes better to new domains, and is effective for multiple few-shot algorithms and network backbones. Code is available at https://github.com/Tsingularity/PoseNorm_Fewshot
Object segmentation in three-dimensional (3-D) point clouds is a critical task for robots capable of 3-D perception. Despite the impressive performance of deep learning-based approaches on object segmentation in 2-D images, deep learning has not been applied nearly as successfully for 3-D point cloud segmentation. Deep networks generally require large amounts of labeled training data, which are readily available for 2-D images but are difficult to produce for 3-D point clouds. In this letter, we present Label Diffusion Lidar Segmentation (LDLS), a novel approach for 3-D point cloud segmentation, which leverages 2-D segmentation of an RGB image from an aligned camera to avoid the need for training on annotated 3-D data. We obtain 2-D segmentation predictions by applying Mask-RCNN to the RGB image, and then link this image to a 3-D lidar point cloud by building a graph of connections among 3-D points and 2-D pixels. This graph then directs a semi-supervised label diffusion process, where the 2-D pixels act as source nodes that diffuse object label information through the 3-D point cloud, resulting in a complete 3-D point cloud segmentation. We conduct empirical studies on the KITTI benchmark dataset and on a mobile robot, demonstrating wide applicability and superior performance of LDLS compared with the previous state of the art in 3-D point cloud segmentation, without any need for either 3-D training data or fine tuning of the 2-D image segmentation model.
We present a technique to improve the generalization of deep representations learned on small labeled datasets by introducing self-supervised tasks as auxiliary loss functions. Although recent research has shown benefits of self-supervised learning (SSL) on large unlabeled datasets, its utility on small datasets is unknown. We find that SSL reduces the relative error rate of few-shot meta-learners by 4%-27%, even when the datasets are small and only utilizing images within the datasets. The improvements are greater when the training set is smaller or the task is more challenging. Though the benefits of SSL may increase with larger training sets, we observe that SSL can have a negative impact on performance when there is a domain shift between distribution of images used for meta-learning and SSL. Based on this analysis we present a technique that automatically select images for SSL from a large, generic pool of unlabeled images for a given dataset using a domain classifier that provides further improvements. We present results using several meta-learners and self-supervised tasks across datasets with varying degrees of domain shifts and label sizes to characterize the effectiveness of SSL for few-shot learning.
In this paper, we present a thorough evaluation of the efficacy of knowledge distillation and its dependence on student and teacher architectures. Starting with the observation that more accurate teachers often don't make good teachers, we attempt to tease apart the factors that affect knowledge distillation performance. We find crucially that larger models do not often make better teachers. We show that this is a consequence of mismatched capacity, and that small students are unable to mimic large teachers. We find typical ways of circumventing this (such as performing a sequence of knowledge distillation steps) to be ineffective. Finally, we show that this effect can be mitigated by stopping the teacher's training early. Our results generalize across datasets and models.
Recent work on single-view 3D reconstruction shows impressive results, but has been restricted to a few fixed categories where extensive training data is available. The problem of generalizing these models to new classes with limited training data is largely open. To address this problem, we present a new model architecture that reframes single-view 3D reconstruction as learnt, category agnostic refinement of a provided, category-specific prior. The provided prior shape for a novel class can be obtained from as few as one 3D shape from this class. Our model can start reconstructing objects from the novel class using this prior without seeing any training image for this class and without any retraining. Our model outperforms category-agnostic baselines and remains competitive with more sophisticated baselines that finetune on the novel categories. Additionally, our network is capable of improving the reconstruction given multiple views despite not being trained on task of multi-view reconstruction.
Understanding fashion styles and trends is of great potential interest to retailers and consumers alike. The photos people upload to social media are a historical and public data source of how people dress across the world and at different times. While we now have tools to automatically recognize the clothing and style attributes of what people are wearing in these photographs, we lack the ability to analyze spatial and temporal trends in these attributes or make predictions about the future. In this paper, we address this need by providing an automatic framework that analyzes large corpora of street imagery to (a) discover and forecast long-term trends of various fashion attributes as well as automatically discovered styles, and (b) identify spatio-temporally localized events that affect what people wear. We show that our framework makes long term trend forecasts that are >20% more accurate than the prior art, and identifies hundreds of socially meaningful events that impact fashion across the globe.
As 3D point clouds become the representation of choice for multiple vision and graphics applications, the ability to synthesize or reconstruct high-resolution, high-fidelity point clouds becomes crucial. Despite the recent success of deep learning models in discriminative tasks of point clouds, generating point clouds remains challenging. This paper proposes a principled probabilistic framework to generate 3D point clouds by modeling them as a distribution of distributions. Specifically, we learn a two-level hierarchy of distributions where the first level is the distribution of shapes and the second level is the distribution of points given a shape. This formulation allows us to both sample shapes and sample an arbitrary number of points from a shape. Our generative model, named PointFlow, learns each level of the distribution with a continuous normalizing flow. The invertibility of normalizing flows enables the computation of the likelihood during training and allows us to train our model in the variational inference framework. Empirically, we demonstrate that PointFlow achieves state-of-the-art performance in point cloud generation. We additionally show that our model can faithfully reconstruct point clouds and learn useful representations in an unsupervised manner. The code will be available at https://github.com/stevenygd/PointFlow.
We present a technique to improve the transferability of deep representations learned on small labeled datasets by introducing self-supervised tasks as auxiliary loss functions. While recent approaches for self-supervised learning have shown the benefits of training on large unlabeled datasets, we find improvements in generalization even on small datasets and when combined with strong supervision. Learning representations with self-supervised losses reduces the relative error rate of a state-of-the-art meta-learner by 5-25% on several few-shot learning benchmarks, as well as off-the-shelf deep networks on standard classification tasks when training from scratch. We find the benefits of self-supervision increase with the difficulty of the task. Our approach utilizes the images within the dataset to construct self-supervised losses and hence is an effective way of learning transferable representations without relying on any external training data.
Detecting objects such as cars and pedestrians in 3D plays an indispensable role in autonomous driving. Existing approaches largely rely on expensive LiDAR sensors for accurate depth information. While recently pseudo-LiDAR has been introduced as a promising alternative, at a much lower cost based solely on stereo images, there is still a notable performance gap. In this paper we provide substantial advances to the pseudo-LiDAR framework through improvements in stereo depth estimation. Concretely, we adapt the stereo network architecture and loss function to be more aligned with accurate depth estimation of far away objects (currently the primary weakness of pseudo-LiDAR). Further, we explore the idea to leverage cheaper but extremely sparse LiDAR sensors, which alone provide insufficient information for 3D detection, to de-bias our depth estimation. We propose a depth-propagation algorithm, guided by the initial depth estimates, to diffuse these few exact measurements across the entire depth map. We show on the KITTI object detection benchmark that our combined approach yields substantial improvements in depth estimation and stereo-based 3D object detection --- outperforming the previous state-of-the-art detection accuracy for far-away objects by 40%. Our code will be publicly available at https://github.com/mileyan/Pseudo_Lidar_V2.
Traditional recognition methods typically require large, artificially-balanced training classes, while few-shot learning methods are tested on artificially small ones. In contrast to both extremes, real world recognition problems exhibit heavy-tailed class distributions, with cluttered scenes and a mix of coarse and fine-grained class distinctions. We show that prior methods designed for few-shot learning do not work out of the box in these challenging conditions, based on a new "meta-iNat" benchmark. We introduce three parameter-free improvements: (a) better training procedures based on adapting cross-validation to meta-learning, (b) novel architectures that localize objects using limited bounding box annotations before classification, and (c) simple parameter-free expansions of the feature space based on bilinear pooling. Together, these improvements double the accuracy of state-of-the-art models on meta-iNat while generalizing to prior benchmarks, complex neural architectures, and settings with substantial domain shift.