Vision-language models have recently shown great potential on many computer vision tasks. Meanwhile, prior work demonstrates prompt tuning designed for vision-language models could acquire superior performance on few-shot image recognition compared to linear probe, a strong baseline. In real-world applications, many few-shot tasks are correlated, particularly in a specialized area. However, such information is ignored by previous work. Inspired by the fact that modeling task relationships by multi-task learning can usually boost performance, we propose a novel method SoftCPT (Soft Context Sharing for Prompt Tuning) to fine-tune pre-trained vision-language models on multiple target few-shot tasks, simultaneously. Specifically, we design a task-shared meta network to generate prompt vector for each task using pre-defined task name together with a learnable meta prompt as input. As such, the prompt vectors of all tasks will be shared in a soft manner. The parameters of this shared meta network as well as the meta prompt vector are tuned on the joint training set of all target tasks. Extensive experiments on three multi-task few-shot datasets show that SoftCPT outperforms the representative single-task prompt tuning method CoOp [78] by a large margin, implying the effectiveness of multi-task learning in vision-language prompt tuning. The source code and data will be made publicly available.
In computer vision, fine-tuning is the de-facto approach to leverage pre-trained vision models to perform downstream tasks. However, deploying it in practice is quite challenging, due to adopting parameter inefficient global update and heavily relying on high-quality downstream data. Recently, prompt-based learning, which adds a task-relevant prompt to adapt the downstream tasks to pre-trained models, has drastically boosted the performance of many natural language downstream tasks. In this work, we extend this notable transfer ability benefited from prompt into vision models as an alternative to fine-tuning. To this end, we propose parameter-efficient Prompt tuning (Pro-tuning) to adapt frozen vision models to various downstream vision tasks. The key to Pro-tuning is prompt-based tuning, i.e., learning task-specific vision prompts for downstream input images with the pre-trained model frozen. By only training a few additional parameters, it can work on diverse CNN-based and Transformer-based architectures. Extensive experiments evidence that Pro-tuning outperforms fine-tuning in a broad range of vision tasks and scenarios, including image classification (generic objects, class imbalance, image corruption, adversarial robustness, and out-of-distribution generalization), and dense prediction tasks such as object detection and semantic segmentation.
Contrastive language-image pretraining has shown great success in learning visual-textual joint representation from web-scale data, demonstrating remarkable "zero-shot" generalization ability for various image tasks. However, how to effectively expand such new language-image pretraining methods to video domains is still an open problem. In this work, we present a simple yet effective approach that adapts the pretrained language-image models to video recognition directly, instead of pretraining a new model from scratch. More concretely, to capture the long-range dependencies of frames along the temporal dimension, we propose a cross-frame attention mechanism that explicitly exchanges information across frames. Such module is lightweight and can be plugged into pretrained language-image models seamlessly. Moreover, we propose a video-specific prompting scheme, which leverages video content information for generating discriminative textual prompts. Extensive experiments demonstrate that our approach is effective and can be generalized to different video recognition scenarios. In particular, under fully-supervised settings, our approach achieves a top-1 accuracy of 87.1% on Kinectics-400, while using 12 times fewer FLOPs compared with Swin-L and ViViT-H. In zero-shot experiments, our approach surpasses the current state-of-the-art methods by +7.6% and +14.9% in terms of top-1 accuracy under two popular protocols. In few-shot scenarios, our approach outperforms previous best methods by +32.1% and +23.1% when the labeled data is extremely limited. Code and models are available at https://aka.ms/X-CLIP
Machine learning systems, especially the methods based on deep learning, enjoy great success in modern computer vision tasks under experimental settings. Generally, these classic deep learning methods are built on the \emph{i.i.d.} assumption, supposing the training and test data are drawn from a similar distribution independently and identically. However, the aforementioned \emph{i.i.d.} assumption is in general unavailable in the real-world scenario, and as a result, leads to sharp performance decay of deep learning algorithms. Behind this, domain shift is one of the primary factors to be blamed. In order to tackle this problem, we propose using \textbf{Po}tential \textbf{E}nergy \textbf{R}anking (PoER) to decouple the object feature and the domain feature (\emph{i.e.,} appearance feature) in given images, promoting the learning of label-discriminative features while filtering out the irrelevant correlations between the objects and the background. PoER helps the neural networks to capture label-related features which contain the domain information first in shallow layers and then distills the label-discriminative representations out progressively, enforcing the neural networks to be aware of the characteristic of objects and background which is vital to the generation of domain-invariant features. PoER reports superior performance on domain generalization benchmarks, improving the average top-1 accuracy by at least 1.20\% compared to the existing methods. Moreover, we use PoER in the ECCV 2022 NICO Challenge\footnote{https://nicochallenge.com}, achieving top place with only a vanilla ResNet-18. The code has been made available at https://github.com/ForeverPs/PoER.
Nowadays, data-driven deep neural models have already shown remarkable progress on Click-through Rate (CTR) prediction. Unfortunately, the effectiveness of such models may fail when there are insufficient data. To handle this issue, researchers often adopt exploration strategies to examine items based on the estimated reward, e.g., UCB or Thompson Sampling. In the context of Exploitation-and-Exploration for CTR prediction, recent studies have attempted to utilize the prediction uncertainty along with model prediction as the reward score. However, we argue that such an approach may make the final ranking score deviate from the original distribution, and thereby affect model performance in the online system. In this paper, we propose a novel exploration method called \textbf{A}dversarial \textbf{G}radient Driven \textbf{E}xploration (AGE). Specifically, we propose a Pseudo-Exploration Module to simulate the gradient updating process, which can approximate the influence of the samples of to-be-explored items for the model. In addition, for better exploration efficiency, we propose an Dynamic Threshold Unit to eliminate the effects of those samples with low potential CTR. The effectiveness of our approach was demonstrated on an open-access academic dataset. Meanwhile, AGE has also been deployed in a real-world display advertising platform and all online metrics have been significantly improved.
The vanilla GAN (Goodfellow et al. 2014) suffers from mode collapse deeply, which usually manifests as that the images generated by generators tend to have a high similarity amongst them, even though their corresponding latent vectors have been very different. In this paper, we introduce a pluggable diversity penalty module (DPM) to alleviate mode collapse of GANs. It reduces the similarity of image pairs in feature space, i.e., if two latent vectors are different, then we enforce the generator to generate two images with different features. The normalized Gram matrix is used to measure the similarity. We compare the proposed method with Unrolled GAN (Metz et al. 2016), BourGAN (Xiao, Zhong, and Zheng 2018), PacGAN (Lin et al. 2018), VEEGAN (Srivastava et al. 2017) and ALI (Dumoulin et al. 2016) on 2D synthetic dataset, and results show that the diversity penalty module can help GAN capture much more modes of the data distribution. Further, in classification tasks, we apply this method as image data augmentation on MNIST, Fashion- MNIST and CIFAR-10, and the classification testing accuracy is improved by 0.24%, 1.34% and 0.52% compared with WGAN GP (Gulrajani et al. 2017), respectively. In domain translation, diversity penalty module can help StarGAN (Choi et al. 2018) generate more accurate attention masks and accelarate the convergence process. Finally, we quantitatively evaluate the proposed method with IS and FID on CelebA, CIFAR-10, MNIST and Fashion-MNIST, and the results suggest GAN with diversity penalty module gets much higher IS and lower FID compared with some SOTA GAN architectures.
Previous methods for skeleton-based gesture recognition mostly arrange the skeleton sequence into a pseudo picture or spatial-temporal graph and apply deep Convolutional Neural Network (CNN) or Graph Convolutional Network (GCN) for feature extraction. Although achieving superior results, these methods have inherent limitations in dynamically capturing local features of interactive hand parts, and the computing efficiency still remains a serious issue. In this work, the self-attention mechanism is introduced to alleviate this problem. Considering the hierarchical structure of hand joints, we propose an efficient hierarchical self-attention network (HAN) for skeleton-based gesture recognition, which is based on pure self-attention without any CNN, RNN or GCN operators. Specifically, the joint self-attention module is used to capture spatial features of fingers, the finger self-attention module is designed to aggregate features of the whole hand. In terms of temporal features, the temporal self-attention module is utilized to capture the temporal dynamics of the fingers and the entire hand. Finally, these features are fused by the fusion self-attention module for gesture classification. Experiments show that our method achieves competitive results on three gesture recognition datasets with much lower computational complexity.
Inspired by the success of deep learning, recent industrial Click-Through Rate (CTR) prediction models have made the transition from traditional shallow approaches to deep approaches. Deep Neural Networks (DNNs) are known for its ability to learn non-linear interactions from raw feature automatically, however, the non-linear feature interaction is learned in an implicit manner. The non-linear interaction may be hard to capture and explicitly model the \textit{co-action} of raw feature is beneficial for CTR prediction. \textit{Co-action} refers to the collective effects of features toward final prediction. In this paper, we argue that current CTR models do not fully explore the potential of feature co-action. We conduct experiments and show that the effect of feature co-action is underestimated seriously. Motivated by our observation, we propose feature Co-Action Network (CAN) to explore the potential of feature co-action. The proposed model can efficiently and effectively capture the feature co-action, which improves the model performance while reduce the storage and computation consumption. Experiment results on public and industrial datasets show that CAN outperforms state-of-the-art CTR models by a large margin. Up to now, CAN has been deployed in the Alibaba display advertisement system, obtaining averaging 12\% improvement on CTR and 8\% on RPM.
Object detectors commonly vary quality according to scales, where the performance on small objects is the least satisfying. In this paper, we investigate this phenomenon and discover that: in the majority of training iterations, small objects contribute barely to the total loss, causing poor performance with imbalanced optimization. Inspired by this finding, we present Stitcher, a feedback-driven data provider, which aims to train object detectors in a balanced way. In Stitcher, images are resized into smaller components and then stitched into the same size to regular images. Stitched images contain inevitable smaller objects, which would be beneficial with our core idea, to exploit the loss statistics as feedback to guide next-iteration update. Experiments have been conducted on various detectors, backbones, training periods, datasets, and even on instance segmentation. Stitcher steadily improves performance by a large margin in all settings, especially for small objects, with nearly no additional computation in both training and testing stages.