arXiv subjects

Jianfei Cai

Publications and source records attributed to Jianfei Cai.

At least 145 records · Page 8Linked to original sources

arXiv2022

ProposalCLIP: Unsupervised Open-Category Object Proposal Generation via Exploiting CLIP Cues

Object proposal generation is an important and fundamental task in computer vision. In this paper, we propose ProposalCLIP, a method towards unsupervised open-category object proposal generation. Unlike previous works which require a large number of bounding box annotations and/or can only generate proposals for limited object categories, our ProposalCLIP is able to predict proposals for a large variety of object categories without annotations, by exploiting CLIP (contrastive language-image pre-training) cues. Firstly, we analyze CLIP for unsupervised open-category proposal generation and design an objectness score based on our empirical analysis on proposal selection. Secondly, a graph-based merging module is proposed to solve the limitations of CLIP cues and merge fragmented proposals. Finally, we present a proposal regression module that extracts pseudo labels based on CLIP cues and trains a lightweight network to further refine proposals. Extensive experiments on PASCAL VOC, COCO and Visual Genome datasets show that our ProposalCLIP can better generate proposals than previous state-of-the-art methods. Our ProposalCLIP also shows benefits for downstream tasks, such as unsupervised object detection.

cs.CV↗

arXiv2022

MED-TEX: Transferring and Explaining Knowledge with Less Data from Pretrained Medical Imaging Models

Deep learning methods usually require a large amount of training data and lack interpretability. In this paper, we propose a novel knowledge distillation and model interpretation framework for medical image classification that jointly solves the above two issues. Specifically, to address the data-hungry issue, a small student model is learned with less data by distilling knowledge from a cumbersome pretrained teacher model. To interpret the teacher model and assist the learning of the student, an explainer module is introduced to highlight the regions of an input that are important for the predictions of the teacher model. Furthermore, the joint framework is trained by a principled way derived from the information-theoretic perspective. Our framework outperforms on the knowledge distillation and model interpretation tasks compared to state-of-the-art methods on a fundus dataset.

cs.CV↗

arXiv2021

Less is More: Pay Less Attention in Vision Transformers

Transformers have become one of the dominant architectures in deep learning, particularly as a powerful alternative to convolutional neural networks (CNNs) in computer vision. However, Transformer training and inference in previous works can be prohibitively expensive due to the quadratic complexity of self-attention over a long sequence of representations, especially for high-resolution dense prediction tasks. To this end, we present a novel Less attention vIsion Transformer (LIT), building upon the fact that the early self-attention layers in Transformers still focus on local patterns and bring minor benefits in recent hierarchical vision Transformers. Specifically, we propose a hierarchical Transformer where we use pure multi-layer perceptrons (MLPs) to encode rich local patterns in the early stages while applying self-attention modules to capture longer dependencies in deeper layers. Moreover, we further propose a learned deformable token merging module to adaptively fuse informative patches in a non-uniform manner. The proposed LIT achieves promising performance on image recognition tasks, including image classification, object detection and instance segmentation, serving as a strong backbone for many vision tasks. Code is available at: https://github.com/zhuang-group/LIT

cs.CV↗

arXiv2021

Bridging Global Context Interactions for High-Fidelity Image Completion

Bridging global context interactions correctly is important for high-fidelity image completion with large masks. Previous methods attempting this via deep or large receptive field (RF) convolutions cannot escape from the dominance of nearby interactions, which may be inferior. In this paper, we propose to treat image completion as a directionless sequence-to-sequence prediction task, and deploy a transformer to directly capture long-range dependence in the encoder. Crucially, we employ a restrictive CNN with small and non-overlapping RF for weighted token representation, which allows the transformer to explicitly model the long-range visible context relations with equal importance in all layers, without implicitly confounding neighboring tokens when larger RFs are used. To improve appearance consistency between visible and generated regions, a novel attention-aware layer (AAL) is introduced to better exploit distantly related high-frequency features. Overall, extensive experiments demonstrate superior performance compared to state-of-the-art methods on several datasets.

cs.CV↗

arXiv2021

High-Resolution Optical Flow from 1D Attention and Correlation

Optical flow is inherently a 2D search problem, and thus the computational complexity grows quadratically with respect to the search window, making large displacements matching infeasible for high-resolution images. In this paper, we take inspiration from Transformers and propose a new method for high-resolution optical flow estimation with significantly less computation. Specifically, a 1D attention operation is first applied in the vertical direction of the target image, and then a simple 1D correlation in the horizontal direction of the attended image is able to achieve 2D correspondence modeling effect. The directions of attention and correlation can also be exchanged, resulting in two 3D cost volumes that are concatenated for optical flow estimation. The novel 1D formulation empowers our method to scale to very high-resolution input images while maintaining competitive performance. Extensive experiments on Sintel, KITTI and real-world 4K ($2160 \times 3840$) resolution images demonstrated the effectiveness and superiority of our proposed method. Code and models are available at \url{https://github.com/haofeixu/flow1d}.

cs.CV↗

arXiv2021

CSG-Stump: A Learning Friendly CSG-Like Representation for Interpretable Shape Parsing

Generating an interpretable and compact representation of 3D shapes from point clouds is an important and challenging problem. This paper presents CSG-Stump Net, an unsupervised end-to-end network for learning shapes from point clouds and discovering the underlying constituent modeling primitives and operations as well. At the core is a three-level structure called {\em CSG-Stump}, consisting of a complement layer at the bottom, an intersection layer in the middle, and a union layer at the top. CSG-Stump is proven to be equivalent to CSG in terms of representation, therefore inheriting the interpretable, compact and editable nature of CSG while freeing from CSG's complex tree structures. Particularly, the CSG-Stump has a simple and regular structure, allowing neural networks to give outputs of a constant dimensionality, which makes itself deep-learning friendly. Due to these characteristics of CSG-Stump, CSG-Stump Net achieves superior results compared to previous CSG-based methods and generates much more appealing shapes, as confirmed by extensive experiments. Project page: https://kimren227.github.io/projects/CSGStump/

cs.CV↗

arXiv2021

Auto-Parsing Network for Image Captioning and Visual Question Answering

We propose an Auto-Parsing Network (APN) to discover and exploit the input data's hidden tree structures for improving the effectiveness of the Transformer-based vision-language systems. Specifically, we impose a Probabilistic Graphical Model (PGM) parameterized by the attention operations on each self-attention layer to incorporate sparse assumption. We use this PGM to softly segment an input sequence into a few clusters where each cluster can be treated as the parent of the inside entities. By stacking these PGM constrained self-attention layers, the clusters in a lower layer compose into a new sequence, and the PGM in a higher layer will further segment this sequence. Iteratively, a sparse tree can be implicitly parsed, and this tree's hierarchical knowledge is incorporated into the transformed embeddings, which can be used for solving the target vision-language tasks. Specifically, we showcase that our APN can strengthen Transformer based networks in two major vision-language tasks: Captioning and Visual Question Answering. Also, a PGM probability-based parsing algorithm is developed by which we can discover what the hidden structure of input is during the inference.

cs.CV↗

arXiv2021

Semantic Compositional Learning for Low-shot Scene Graph Generation

Scene graphs provide valuable information to many downstream tasks. Many scene graph generation (SGG) models solely use the limited annotated relation triples for training, leading to their underperformance on low-shot (few and zero) scenarios, especially on the rare predicates. To address this problem, we propose a novel semantic compositional learning strategy that makes it possible to construct additional, realistic relation triples with objects from different images. Specifically, our strategy decomposes a relation triple by identifying and removing the unessential component and composes a new relation triple by fusing with a semantically or visually similar object from a visual components dictionary, whilst ensuring the realisticity of the newly composed triple. Notably, our strategy is generic and can be combined with existing SGG models to significantly improve their performance. We performed a comprehensive evaluation on the benchmark dataset Visual Genome. For three recent SGG models, adding our strategy improves their performance by close to 50\%, and all of them substantially exceed the current state-of-the-art.

cs.CV↗

arXiv2021

Scalable Vision Transformers with Hierarchical Pooling

The recently proposed Visual image Transformers (ViT) with pure attention have achieved promising performance on image recognition tasks, such as image classification. However, the routine of the current ViT model is to maintain a full-length patch sequence during inference, which is redundant and lacks hierarchical representation. To this end, we propose a Hierarchical Visual Transformer (HVT) which progressively pools visual tokens to shrink the sequence length and hence reduces the computational cost, analogous to the feature maps downsampling in Convolutional Neural Networks (CNNs). It brings a great benefit that we can increase the model capacity by scaling dimensions of depth/width/resolution/patch size without introducing extra computational complexity due to the reduced sequence length. Moreover, we empirically find that the average pooled visual tokens contain more discriminative information than the single class token. To demonstrate the improved scalability of our HVT, we conduct extensive experiments on the image classification task. With comparable FLOPs, our HVT outperforms the competitive baselines on ImageNet and CIFAR-100 datasets. Code is available at https://github.com/MonashAI/HVT

cs.CV↗

arXiv2021

Learning Meta-class Memory for Few-Shot Semantic Segmentation

Currently, the state-of-the-art methods treat few-shot semantic segmentation task as a conditional foreground-background segmentation problem, assuming each class is independent. In this paper, we introduce the concept of meta-class, which is the meta information (e.g. certain middle-level features) shareable among all classes. To explicitly learn meta-class representations in few-shot segmentation task, we propose a novel Meta-class Memory based few-shot segmentation method (MM-Net), where we introduce a set of learnable memory embeddings to memorize the meta-class information during the base class training and transfer to novel classes during the inference stage. Moreover, for the $k$-shot scenario, we propose a novel image quality measurement module to select images from the set of support images. A high-quality class prototype could be obtained with the weighted sum of support image features based on the quality measure. Experiments on both PASCAL-$5^i$ and COCO dataset shows that our proposed method is able to achieve state-of-the-art results in both 1-shot and 5-shot settings. Particularly, our proposed MM-Net achieves 37.5\% mIoU on the COCO dataset in 1-shot setting, which is 5.1\% higher than the previous state-of-the-art.

cs.CV↗

arXiv2021

Remember What You have drawn: Semantic Image Manipulation with Memory

Image manipulation with natural language, which aims to manipulate images with the guidance of language descriptions, has been a challenging problem in the fields of computer vision and natural language processing (NLP). Currently, a number of efforts have been made for this task, but their performances are still distant away from generating realistic and text-conformed manipulated images. Therefore, in this paper, we propose a memory-based Image Manipulation Network (MIM-Net), where a set of memories learned from images is introduced to synthesize the texture information with the guidance of the textual description. We propose a two-stage network with an additional reconstruction stage to learn the latent memories efficiently. To avoid the unnecessary background changes, we propose a Target Localization Unit (TLU) to focus on the manipulation of the region mentioned by the text. Moreover, to learn a robust memory, we further propose a novel randomized memory training loss. Experiments on the four popular datasets show the better performance of our method compared to the existing ones.

cs.CV↗

arXiv2021

Multi-Label Image Classification with Contrastive Learning

Recently, as an effective way of learning latent representations, contrastive learning has been increasingly popular and successful in various domains. The success of constrastive learning in single-label classifications motivates us to leverage this learning framework to enhance distinctiveness for better performance in multi-label image classification. In this paper, we show that a direct application of contrastive learning can hardly improve in multi-label cases. Accordingly, we propose a novel framework for multi-label classification with contrastive learning in a fully supervised setting, which learns multiple representations of an image under the context of different labels. This facilities a simple yet intuitive adaption of contrastive learning into our model to boost its performance in multi-label image classification. Extensive experiments on two benchmark datasets show that the proposed framework achieves state-of-the-art performance in the comparison with the advanced methods in multi-label classification.

cs.CV↗

arXiv2021

Semi-supervised Left Atrium Segmentation with Mutual Consistency Training

Semi-supervised learning has attracted great attention in the field of machine learning, especially for medical image segmentation tasks, since it alleviates the heavy burden of collecting abundant densely annotated data for training. However, most of existing methods underestimate the importance of challenging regions (e.g. small branches or blurred edges) during training. We believe that these unlabeled regions may contain more crucial information to minimize the uncertainty prediction for the model and should be emphasized in the training process. Therefore, in this paper, we propose a novel Mutual Consistency Network (MC-Net) for semi-supervised left atrium segmentation from 3D MR images. Particularly, our MC-Net consists of one encoder and two slightly different decoders, and the prediction discrepancies of two decoders are transformed as an unsupervised loss by our designed cycled pseudo label scheme to encourage mutual consistency. Such mutual consistency encourages the two decoders to have consistent and low-entropy predictions and enables the model to gradually capture generalized features from these unlabeled challenging regions. We evaluate our MC-Net on the public Left Atrium (LA) database and it obtains impressive performance gains by exploiting the unlabeled data effectively. Our MC-Net outperforms six recent semi-supervised methods for left atrium segmentation, and sets the new state-of-the-art performance on the LA database.

cs.CV↗

arXiv2021

Half-body Portrait Relighting with Overcomplete Lighting Representation

We present a neural-based model for relighting a half-body portrait image by simply referring to another portrait image with the desired lighting condition. Rather than following classical inverse rendering methodology that involves estimating normals, albedo and environment maps, we implicitly encode the subject and lighting in a latent space, and use these latent codes to generate relighted images by neural rendering. A key technical innovation is the use of a novel overcomplete lighting representation, which facilitates lighting interpolation in the latent space, as well as helping regularize the self-organization of the lighting latent space during training. In addition, we propose a novel multiplicative neural render that more effectively combines the subject and lighting latent codes for rendering. We also created a large-scale photorealistic rendered relighting dataset for training, which allows our model to generalize well to real images. Extensive experiments demonstrate that our system not only outperforms existing methods for referral-based portrait relighting, but also has the capability generate sequences of relighted images via lighting rotations.

cs.GR↗

arXiv2021

Unconstrained Facial Action Unit Detection via Latent Feature Domain

Facial action unit (AU) detection in the wild is a challenging problem, due to the unconstrained variability in facial appearances and the lack of accurate annotations. Most existing methods depend on either impractical labor-intensive labeling or inaccurate pseudo labels. In this paper, we propose an end-to-end unconstrained facial AU detection framework based on domain adaptation, which transfers accurate AU labels from a constrained source domain to an unconstrained target domain by exploiting labels of AU-related facial landmarks. Specifically, we map a source image with label and a target image without label into a latent feature domain by combining source landmark-related feature with target landmark-free feature. Due to the combination of source AU-related information and target AU-free information, the latent feature domain with transferred source label can be learned by maximizing the target-domain AU detection performance. Moreover, we introduce a novel landmark adversarial loss to disentangle the landmark-free feature from the landmark-related feature by treating the adversarial learning as a multi-player minimax game. Our framework can also be naturally extended for use with target-domain pseudo AU labels. Extensive experiments show that our method soundly outperforms lower-bounds and upper-bounds of the basic model, as well as state-of-the-art approaches on the challenging in-the-wild benchmarks. The code is available at https://github.com/ZhiwenShao/ADLD.

cs.CV↗

arXiv2021

RSG: A Simple but Effective Module for Learning Imbalanced Datasets

Imbalanced datasets widely exist in practice and area great challenge for training deep neural models with agood generalization on infrequent classes. In this work, wepropose a new rare-class sample generator (RSG) to solvethis problem. RSG aims to generate some new samplesfor rare classes during training, and it has in particularthe following advantages: (1) it is convenient to use andhighly versatile, because it can be easily integrated intoany kind of convolutional neural network, and it works wellwhen combined with different loss functions, and (2) it isonly used during the training phase, and therefore, no ad-ditional burden is imposed on deep neural networks duringthe testing phase. In extensive experimental evaluations, weverify the effectiveness of RSG. Furthermore, by leveragingRSG, we obtain competitive results on Imbalanced CIFARand new state-of-the-art results on Places-LT, ImageNet-LT, and iNaturalist 2018. The source code is available at https://github.com/Jianf-Wang/RSG.

cs.CV↗

arXiv2021

Visiting the Invisible: Layer-by-Layer Completed Scene Decomposition

Existing scene understanding systems mainly focus on recognizing the visible parts of a scene, ignoring the intact appearance of physical objects in the real-world. Concurrently, image completion has aimed to create plausible appearance for the invisible regions, but requires a manual mask as input. In this work, we propose a higher-level scene understanding system to tackle both visible and invisible parts of objects and backgrounds in a given scene. Particularly, we built a system to decompose a scene into individual objects, infer their underlying occlusion relationships, and even automatically learn which parts of the objects are occluded that need to be completed. In order to disentangle the occluded relationships of all objects in a complex scene, we use the fact that the front object without being occluded is easy to be identified, detected, and segmented. Our system interleaves the two tasks of instance segmentation and scene completion through multiple iterations, solving for objects layer-by-layer. We first provide a thorough experiment using a new realistically rendered dataset with ground-truths for all invisible regions. To bridge the domain gap to real imagery where ground-truths are unavailable, we then train another model with the pseudo-ground-truths generated from our trained synthesis model. We demonstrate results on a wide variety of datasets and show significant improvement over the state-of-the-art.

cs.CV↗

arXiv2021

The Spatially-Correlative Loss for Various Image Translation Tasks

We propose a novel spatially-correlative loss that is simple, efficient and yet effective for preserving scene structure consistency while supporting large appearance changes during unpaired image-to-image (I2I) translation. Previous methods attempt this by using pixel-level cycle-consistency or feature-level matching losses, but the domain-specific nature of these losses hinder translation across large domain gaps. To address this, we exploit the spatial patterns of self-similarity as a means of defining scene structure. Our spatially-correlative loss is geared towards only capturing spatial relationships within an image rather than domain appearance. We also introduce a new self-supervised learning method to explicitly learn spatially-correlative maps for each specific translation task. We show distinct improvement over baseline models in all three modes of unpaired I2I translation: single-modal, multi-modal, and even single-image translation. This new loss can easily be integrated into existing network architectures and thus allows wide applicability.

cs.CV↗