Hao Yu

Zheng Qin, Hao Yu, Changjian Wang et al.

We study the problem of outlier correspondence pruning for non-rigid point cloud registration. In rigid registration, spatial consistency has been a commonly used criterion to discriminate outliers from inliers. It measures the compatibility of two correspondences by the discrepancy between the respective distances in two point clouds. However, spatial consistency no longer holds in non-rigid cases and outlier rejection for non-rigid registration has not been well studied. In this work, we propose Graph-based Spatial Consistency Network (GraphSCNet) to filter outliers for non-rigid registration. Our method is based on the fact that non-rigid deformations are usually locally rigid, or local shape preserving. We first design a local spatial consistency measure over the deformation graph of the point cloud, which evaluates the spatial compatibility only between the correspondences in the vicinity of a graph node. An attention-based non-rigid correspondence embedding module is then devised to learn a robust representation of non-rigid correspondences from local spatial consistency. Despite its simplicity, GraphSCNet effectively improves the quality of the putative correspondences and attains state-of-the-art performance on three challenging benchmarks. Our code and models are available at https://github.com/qinzheng93/GraphSCNet.

Chuhui Xue, Wenqing Zhang, Yu Hao et al.

Recently, Vision-Language Pre-training (VLP) techniques have greatly benefited various vision-language tasks by jointly learning visual and textual representations, which intuitively helps in Optical Character Recognition (OCR) tasks due to the rich visual and textual information in scene text images. However, these methods cannot well cope with OCR tasks because of the difficulty in both instance-level text encoding and image-text pair acquisition (i.e. images and captured texts in them). This paper presents a weakly supervised pre-training method, oCLIP, which can acquire effective scene text representations by jointly learning and aligning visual and textual information. Our network consists of an image encoder and a character-aware text encoder that extract visual and textual features, respectively, as well as a visual-textual decoder that models the interaction among textual and visual features for learning effective scene text representations. With the learning of textual features, the pre-trained model can attend texts in images well with character awareness. Besides, these designs enable the learning from weakly annotated texts (i.e. partial texts in images without text bounding boxes) which mitigates the data annotation constraint greatly. Experiments over the weakly annotated images in ICDAR2019-LSVT show that our pre-trained model improves F-score by +2.5\% and +4.8\% while transferring its weights to other text detection and spotting networks, respectively. In addition, the proposed method outperforms existing pre-training techniques consistently across multiple public datasets (e.g., +3.2\% and +1.3\% for Total-Text and CTW1500).

Fenia Christopoulou, Gerasimos Lampouras, Milan Gritta et al.

We present PanGu-Coder, a pretrained decoder-only language model adopting the PanGu-Alpha architecture for text-to-code generation, i.e. the synthesis of programming language solutions given a natural language problem description. We train PanGu-Coder using a two-stage strategy: the first stage employs Causal Language Modelling (CLM) to pre-train on raw programming language data, while the second stage uses a combination of Causal Language Modelling and Masked Language Modelling (MLM) training objectives that focus on the downstream task of text-to-code generation and train on loosely curated pairs of natural language program definitions and code functions. Finally, we discuss PanGu-Coder-FT, which is fine-tuned on a combination of competitive programming problems and code with continuous integration tests. We evaluate PanGu-Coder with a focus on whether it generates functionally correct programs and demonstrate that it achieves equivalent or better performance than similarly sized models, such as CodeX, while attending a smaller context window and training on less data.

Yu Hao, Haoyang Pei, Yixuan Lyu et al.

Deep learning has made great strides for object detection in images. The detection accuracy and computational cost of object detection depend on the spatial resolution of an image, which may be constrained by both the camera and storage considerations. Compression is often achieved by reducing either spatial or amplitude resolution or, at times, both, both of which have well-known effects on performance. Detection accuracy also depends on the distance of the object of interest from the camera. Our work examines the impact of spatial and amplitude resolution, as well as object distance, on object detection accuracy and computational cost. We develop a resolution-adaptive variant of YOLOv5 (RA-YOLO), which varies the number of scales in the feature pyramid and detection head based on the spatial resolution of the input image. To train and evaluate this new method, we created a dataset of images with diverse spatial and amplitude resolutions by combining images from the TJU and Eurocity datasets and generating different resolutions by applying spatial resizing and compression. We first show that RA-YOLO achieves a good trade-off between detection accuracy and inference time over a large range of spatial resolutions. We then evaluate the impact of spatial and amplitude resolutions on object detection accuracy using the proposed RA-YOLO model. We demonstrate that the optimal spatial resolution that leads to the highest detection accuracy depends on the 'tolerated' image size. We further assess the impact of the distance of an object to the camera on the detection accuracy and show that higher spatial resolution enables a greater detection range. These results provide important guidelines for choosing the image spatial resolution and compression settings predicated on available bandwidth, storage, desired inference time, and/or desired detection range, in practical applications.

Jiayuan Zhuang, Zheng Qin, Hao Yu et al.

Classification and localization are two main sub-tasks in object detection. Nonetheless, these two tasks have inconsistent preferences for feature context, i.e., localization expects more boundary-aware features to accurately regress the bounding box, while more semantic context is preferred for object classification. Exsiting methods usually leverage disentangled heads to learn different feature context for each task. However, the heads are still applied on the same input features, which leads to an imperfect balance between classifcation and localization. In this work, we propose a novel Task-Specific COntext DEcoupling (TSCODE) head which further disentangles the feature encoding for two tasks. For classification, we generate spatially-coarse but semantically-strong feature encoding. For localization, we provide high-resolution feature map containing more edge information to better regress object boundaries. TSCODE is plug-and-play and can be easily incorperated into existing detection pipelines. Extensive experiments demonstrate that our method stably improves different detectors by over 1.0 AP with less computational cost. Our code and models will be publicly released.

Binbin Li, Xinyu Du, Yao Hu et al.

Online camera-to-ground calibration is to generate a non-rigid body transformation between the camera and the road surface in a real-time manner. Existing solutions utilize static calibration, suffering from environmental variations such as tire pressure changes, vehicle loading volume variations, and road surface diversity. Other online solutions exploit the usage of road elements or photometric consistency between overlapping views across images, which require continuous detection of specific targets on the road or assistance with multiple cameras to facilitate calibration. In our work, we propose an online monocular camera-to-ground calibration solution that does not utilize any specific targets while driving. We perform a coarse-to-fine approach for ground feature extraction through wheel odometry and estimate the camera-to-ground calibration parameters through a sliding-window-based factor graph optimization. Considering the non-rigid transformation of camera-to-ground while driving, we provide metrics to quantify calibration performance and stopping criteria to report/broadcast our satisfying calibration results. Extensive experiments using real-world data demonstrate that our algorithm is effective and outperforms state-of-the-art techniques.

Zhangzi Zhu, Yu Hao, Wenqing Zhang et al.

This report presents our 2nd place solution to ECCV 2022 challenge on Out-of-Vocabulary Scene Text Understanding (OOV-ST) : Cropped Word Recognition. This challenge is held in the context of ECCV 2022 workshop on Text in Everything (TiE), which aims to extract out-of-vocabulary words from natural scene images. In the competition, we first pre-train SCATTER on the synthetic datasets, then fine-tune the model on the training set with data augmentations. Meanwhile, two additional models are trained specifically for long and vertical texts. Finally, we combine the output from different models with different layers, different backbones, and different seeds as the final results. Our solution achieves a word accuracy of 59.45\% when considering out-of-vocabulary words only.

Zhangzi Zhu, Chuhui Xue, Yu Hao et al.

Scene text recognition has attracted increasing interest in recent years due to its wide range of applications in multilingual translation, autonomous driving, etc. In this report, we describe our solution to the Out of Vocabulary Scene Text Understanding (OOV-ST) Challenge, which aims to extract out-of-vocabulary (OOV) words from natural scene images. Our oCLIP-based model achieves 28.59\% in h-mean which ranks 1st in end-to-end OOV word recognition track of OOV Challenge in ECCV2022 TiE Workshop.

Yun-Hao Cao, Hao Yu, Jianxin Wu

Vision Transformers (ViTs) is emerging as an alternative to convolutional neural networks (CNNs) for visual recognition. They achieve competitive results with CNNs but the lack of the typical convolutional inductive bias makes them more data-hungry than common CNNs. They are often pretrained on JFT-300M or at least ImageNet and few works study training ViTs with limited data. In this paper, we investigate how to train ViTs with limited data (e.g., 2040 images). We give theoretical analyses that our method (based on parametric instance discrimination) is superior to other methods in that it can capture both feature alignment and instance similarities. We achieve state-of-the-art results when training from scratch on 7 small datasets under various ViT backbones. We also investigate the transferring ability of small datasets and find that representations learned from small datasets can even improve large-scale ImageNet training.

Nataniel Ruiz, Hao Yu, Danielle A. Allessio et al.

In this work, we propose a video-based transfer learning approach for predicting problem outcomes of students working with an intelligent tutoring system (ITS). By analyzing a student's face and gestures, our method predicts the outcome of a student answering a problem in an ITS from a video feed. Our work is motivated by the reasoning that the ability to predict such outcomes enables tutoring systems to adjust interventions, such as hints and encouragement, and to ultimately yield improved student learning. We collected a large labeled dataset of student interactions with an intelligent online math tutor consisting of 68 sessions, where 54 individual students solved 2,749 problems. The dataset is public and available at https://www.cs.bu.edu/faculty/betke/research/learning/ . Working with this dataset, our transfer-learning challenge was to design a representation in the source domain of pictures obtained "in the wild" for the task of facial expression analysis, and transferring this learned representation to the task of human behavior prediction in the domain of webcam videos of students in a classroom environment. We developed a novel facial affect representation and a user-personalized training scheme that unlocks the potential of this representation. We designed several variants of a recurrent neural network that models the temporal structure of video sequences of students solving math problems. Our final model, named ATL-BP for Affect Transfer Learning for Behavior Prediction, achieves a relative increase in mean F-score of 50% over the state-of-the-art method on this new dataset.