Visual object tracking (VOT) is an essential component for many applications, such as autonomous driving or assistive robotics. However, recent works tend to develop accurate systems based on more computationally expensive feature extractors for bett
er instance matching. In contrast, this work addresses the importance of motion prediction in VOT. We use an off-the-shelf object detector to obtain instance bounding boxes. Then, a combination of camera motion decouple and Kalman filter is used for state estimation. Although our baseline system is a straightforward combination of standard methods, we obtain state-of-the-art results. Our method establishes new state-of-the-art performance on VOT (VOT-2016 and VOT-2018). Our proposed method improves the EAO on VOT-2016 from 0.472 of prior art to 0.505, from 0.410 to 0.431 on VOT-2018. To show the generalizability, we also test our method on video object segmentation (VOS: DAVIS-2016 and DAVIS-2017) and observe consistent improvement.
A common approach to localize 3D human joints in a synchronized and calibrated multi-view setup consists of two-steps: (1) apply a 2D detector separately on each view to localize joints in 2D, and (2) perform robust triangulation on 2D detections fro
m each view to acquire the 3D joint locations. However, in step 1, the 2D detector is limited to solving challenging cases which could potentially be better resolved in 3D, such as occlusions and oblique viewing angles, purely in 2D without leveraging any 3D information. Therefore, we propose the differentiable epipolar transformer, which enables the 2D detector to leverage 3D-aware features to improve 2D pose estimation. The intuition is: given a 2D location p in the current view, we would like to first find its corresponding point p in a neighboring view, and then combine the features at p with the features at p, thus leading to a 3D-aware feature at p. Inspired by stereo matching, the epipolar transformer leverages epipolar constraints and feature matching to approximate the features at p. Experiments on InterHand and Human3.6M show that our approach has consistent improvements over the baselines. Specifically, in the condition where no external data is used, our Human3.6M model trained with ResNet-50 backbone and image size 256 x 256 outperforms state-of-the-art by 4.23 mm and achieves MPJPE 26.9 mm.
Mistakes/uncertainties in object detection could lead to catastrophes when deploying robots in the real world. In this paper, we measure the uncertainties of object localization to minimize this kind of risk. Uncertainties emerge upon challenging cas
es like occlusion. The bounding box borders of an occluded object can have multiple plausible configurations. We propose a deep multivariate mixture of Gaussians model for probabilistic object detection. The covariances help to learn the relationship between the borders, and the mixture components potentially learn different configurations of an occluded part. Quantitatively, our model improves the AP of the baselines by 3.9% and 1.4% on CrowdHuman and MS-COCO respectively with almost no computational or memory overhead. Qualitatively, our model enjoys explainability since the resulting covariance matrices and the mixture components help measure uncertainties.
Very deep convolutional neural networks (CNNs) have been firmly established as the primary methods for many computer vision tasks. However, most state-of-the-art CNNs are large, which results in high inference latency. Recently, depth-wise separable
convolution has been proposed for image recognition tasks on computationally limited platforms such as robotics and self-driving cars. Though it is much faster than its counterpart, regular convolution, accuracy is sacrificed. In this paper, we propose a novel decomposition approach based on SVD, namely depth-wise decomposition, for expanding regular convolutions into depthwise separable convolutions while maintaining high accuracy. We show our approach can be further generalized to the multi-channel and multi-layer cases, based on Generalized Singular Value Decomposition (GSVD) [59]. We conduct thorough experiments with the latest ShuffleNet V2 model [47] on both random synthesized dataset and a large-scale image recognition dataset: ImageNet [10]. Our approach outperforms channel decomposition [73] on all datasets. More importantly, our approach improves the Top-1 accuracy of ShuffleNet V2 by ~2%.
MobileNet and Binary Neural Networks are two among the most widely used techniques to construct deep learning models for performing a variety of tasks on mobile and embedded platforms.In this paper, we present a simple yet efficient scheme to exploit
MobileNet binarization at activation function and model weights. However, training a binary network from scratch with separable depth-wise and point-wise convolutions in case of MobileNet is not trivial and prone to divergence. To tackle this training issue, we propose a novel neural network architecture, namely MoBiNet - Mobile Binary Network in which skip connections are manipulated to prevent information loss and vanishing gradient, thus facilitate the training process. More importantly, while existing binary neural networks often make use of cumbersome backbones such as Alex-Net, ResNet, VGG-16 with float-type pre-trained weights initialization, our MoBiNet focuses on binarizing the already-compressed neural networks like MobileNet without the need of a pre-trained model to start with. Therefore, our proposal results in an effectively small model while keeping the accuracy comparable to existing ones. Experiments on ImageNet dataset show the potential of the MoBiNet as it achieves 54.40% top-1 accuracy and dramatically reduces the computational cost with binary operators.
We introduce a prediction driven method for visual tracking and segmentation in videos. Instead of solely relying on matching with appearance cues for tracking, we build a predictive model which guides finding more accurate tracking regions efficient
ly. With the proposed prediction mechanism, we improve the model robustness against distractions and occlusions during tracking. We demonstrate significant improvements over state-of-the-art methods not only on visual tracking tasks (VOT 2016 and VOT 2018) but also on video segmentation datasets (DAVIS 2016 and DAVIS 2017).
We motivate and present feature selective anchor-free (FSAF) module, a simple and effective building block for single-shot object detectors. It can be plugged into single-shot detectors with feature pyramid structure. The FSAF module addresses two li
mitations brought up by the conventional anchor-based detection: 1) heuristic-guided feature selection; 2) overlap-based anchor sampling. The general concept of the FSAF module is online feature selection applied to the training of multi-level anchor-free branches. Specifically, an anchor-free branch is attached to each level of the feature pyramid, allowing box encoding and decoding in the anchor-free manner at an arbitrary level. During training, we dynamically assign each instance to the most suitable feature level. At the time of inference, the FSAF module can work jointly with anchor-based branches by outputting predictions in parallel. We instantiate this concept with simple implementations of anchor-free branches and online feature selection strategy. Experimental results on the COCO detection track show that our FSAF module performs better than anchor-based counterparts while being faster. When working jointly with anchor-based branches, the FSAF module robustly improves the baseline RetinaNet by a large margin under various settings, while introducing nearly free inference overhead. And the resulting best model can achieve a state-of-the-art 44.6% mAP, outperforming all existing single-shot detectors on COCO.
In this project, we worked on speech recognition, specifically predicting individual words based on both the video frames and audio. Empowered by convolutional neural networks, the recent speech recognition and lip reading models are comparable to hu
man level performance. We re-implemented and made derivations of the state-of-the-art model. Then, we conducted rich experiments including the effectiveness of attention mechanism, more accurate residual network as the backbone with pre-trained weights and the sensitivity of our model with respect to audio input with/without noise.
Large-scale object detection datasets (e.g., MS-COCO) try to define the ground truth bounding boxes as clear as possible. However, we observe that ambiguities are still introduced when labeling the bounding boxes. In this paper, we propose a novel bo
unding box regression loss for learning bounding box transformation and localization variance together. Our loss greatly improves the localization accuracies of various architectures with nearly no additional computation. The learned localization variance allows us to merge neighboring bounding boxes during non-maximum suppression (NMS), which further improves the localization performance. On MS-COCO, we boost the Average Precision (AP) of VGG-16 Faster R-CNN from 23.6% to 29.1%. More importantly, for ResNet-50-FPN Mask R-CNN, our method improves the AP and AP90 by 1.8% and 6.2% respectively, which significantly outperforms previous state-of-the-art bounding box refinement methods. Our code and models are available at: github.com/yihui-he/KL-Loss
We propose a collection of three shift-based primitives for building efficient compact CNN-based networks. These three primitives (channel shift, address shift, shortcut shift) can reduce the inference time on GPU while maintains the prediction accur
acy. These shift-based primitives only moves the pointer but avoids memory copy, thus very fast. For example, the channel shift operation is 12.7x faster compared to channel shuffle in ShuffleNet but achieves the same accuracy. The address shift and channel shift can be merged into the point-wise group convolution and invokes only a single kernel call, taking little time to perform spatial convolution and channel shift. Shortcut shift requires no time to realize residual connection through allocating space in advance. We blend these shift-based primitives with point-wise group convolution and built two inference-efficient CNN architectures named AddressNet and Enhanced AddressNet. Experiments on CIFAR100 and ImageNet datasets show that our models are faster and achieve comparable or better accuracy.
