Cong Yu

Junru Wu, Yi Liang, Feng Han et al.

Self-supervised pre-training recently demonstrates success on large-scale multimodal data, and state-of-the-art contrastive learning methods often enforce the feature consistency from cross-modality inputs, such as video/audio or video/text pairs. Despite its convenience to formulate and leverage in practice, such cross-modality alignment (CMA) is only a weak and noisy supervision, since two modalities can be semantically misaligned even they are temporally aligned. For example, even in the commonly adopted instructional videos, a speaker can sometimes refer to something that is not visually present in the current frame; and the semantic misalignment would only be more unpredictable for the raw videos from the internet. We conjecture that might cause conflicts and biases among modalities, and may hence prohibit CMA from scaling up to training with larger and more heterogeneous data. This paper first verifies our conjecture by observing that, even in the latest VATT pre-training using only instructional videos, there exist strong gradient conflicts between different CMA losses within the same video, audio, text triplet, indicating them as the noisy source of supervision. We then propose to harmonize such gradients, via two techniques: (i) cross-modality gradient realignment: modifying different CMA loss gradients for each sample triplet, so that their gradient directions are more aligned; and (ii) gradient-based curriculum learning: leveraging the gradient conflict information on an indicator of sample noisiness, to develop a curriculum learning strategy to prioritize training on less noisy sample triplets. Applying those techniques to pre-training VATT on the HowTo100M dataset, we consistently improve its performance on different downstream tasks. Moreover, we are able to scale VATT pre-training to more complicated non-narrative Youtube8M dataset to further improve the state-of-the-arts.

Cong Yu, Dongheng Zhang, Zhi Wu et al.

This paper introduces a novel framework, i.e., RFPose-OT, to enable the 3D human pose estimation from Radio Frequency (RF) signals. Different from existing methods that predict human poses from RF signals on the signal level directly, we consider the structure difference between the RF signals and the human poses, propose to transform the RF signals to the pose domain on the feature level based on Optimal Transport (OT) theory, and generate human poses from the transformed features. To evaluate RFPose-OT, we build a radio system and a multi-view camera system to acquire the RF signal data and the ground-truth human poses. The experimental results in basic indoor environment, occlusion indoor environment, and outdoor environment, all demonstrate that RFPose-OT can predict 3D human poses with higher precision than the state-of-the-art methods.

Jiaxin Huang, Tianqi Liu, Jialu Liu et al.

Multi-Task Learning (MTL) models have shown their robustness, effectiveness, and efficiency for transferring learned knowledge across tasks. In real industrial applications such as web content classification, multiple classification tasks are predicted from the same input text such as a web article. However, at the serving time, the existing multitask transformer models such as prompt or adaptor based approaches need to conduct N forward passes for N tasks with O(N) computation cost. To tackle this problem, we propose a scalable method that can achieve stronger performance with close to O(1) computation cost via only one forward pass. To illustrate real application usage, we release a multitask dataset on news topic and style classification. Our experiments show that our proposed method outperforms strong baselines on both the GLUE benchmark and our news dataset. Our code and dataset are publicly available at https://bit.ly/mtop-code.

Donghui Hu, Yu Zhang, Cong Yu et al.

Image steganography is the art and science of using images as cover for covert communications. With the development of neural networks, traditional image steganography is more likely to be detected by deep learning-based steganalysis. To improve upon this, we propose image steganography network based on style transfer, and the embedding of secret messages can be disguised as image stylization. We embed secret information while transforming the content image style. In latent space, the secret information is integrated into the latent representation of the cover image to generate the stego images, which are indistinguishable from normal stylized images. It is an end-to-end unsupervised model without pre-training. Extensive experiments on the benchmark dataset demonstrate the reliability, quality and security of stego images generated by our steganographic network.

Zhi Wu, Dongheng Zhang, Chunyang Xie et al.

Human silhouette segmentation, which is originally defined in computer vision, has achieved promising results for understanding human activities. However, the physical limitation makes existing systems based on optical cameras suffer from severe performance degradation under low illumination, smoke, and/or opaque obstruction conditions. To overcome such limitations, in this paper, we propose to utilize the radio signals, which can traverse obstacles and are unaffected by the lighting conditions to achieve silhouette segmentation. The proposed RFMask framework is composed of three modules. It first transforms RF signals captured by millimeter wave radar on two planes into spatial domain and suppress interference with the signal processing module. Then, it locates human reflections on RF frames and extract features from surrounding signals with human detection module. Finally, the extracted features from RF frames are aggregated with an attention based mask generation module. To verify our proposed framework, we collect a dataset containing 804,760 radio frames and 402,380 camera frames with human activities under various scenes. Experimental results show that the proposed framework can achieve impressive human silhouette segmentation even under the challenging scenarios(such as low light and occlusion scenarios) where traditional optical-camera-based methods fail. To the best of our knowledge, this is the first investigation towards segmenting human silhouette based on millimeter wave signals. We hope that our work can serve as a baseline and inspire further research that perform vision tasks with radio signals. The dataset and codes will be made in public.

Cong Yu, Zhi Wu, Dongheng Zhang et al.

This paper demonstrates human synthesis based on the Radio Frequency (RF) signals, which leverages the fact that RF signals can record human movements with the signal reflections off the human body. Different from existing RF sensing works that can only perceive humans roughly, this paper aims to generate fine-grained optical human images by introducing a novel cross-modal RFGAN model. Specifically, we first build a radio system equipped with horizontal and vertical antenna arrays to transceive RF signals. Since the reflected RF signals are processed as obscure signal projection heatmaps on the horizontal and vertical planes, we design a RF-Extractor with RNN in RFGAN for RF heatmap encoding and combining to obtain the human activity information. Then we inject the information extracted by the RF-Extractor and RNN as the condition into GAN using the proposed RF-based adaptive normalizations. Finally, we train the whole model in an end-to-end manner. To evaluate our proposed model, we create two cross-modal datasets (RF-Walk & RF-Activity) that contain thousands of optical human activity frames and corresponding RF signals. Experimental results show that the RFGAN can generate target human activity frames using RF signals. To the best of our knowledge, this is the first work to generate optical images based on RF signals.