Huan Yang

Hongwei Xue, Tiankai Hang, Yanhong Zeng et al.

We study joint video and language (VL) pre-training to enable cross-modality learning and benefit plentiful downstream VL tasks. Existing works either extract low-quality video features or learn limited text embedding, while neglecting that high-resolution videos and diversified semantics can significantly improve cross-modality learning. In this paper, we propose a novel High-resolution and Diversified VIdeo-LAnguage pre-training model (HD-VILA) for many visual tasks. In particular, we collect a large dataset with two distinct properties: 1) the first high-resolution dataset including 371.5k hours of 720p videos, and 2) the most diversified dataset covering 15 popular YouTube categories. To enable VL pre-training, we jointly optimize the HD-VILA model by a hybrid Transformer that learns rich spatiotemporal features, and a multimodal Transformer that enforces interactions of the learned video features with diversified texts. Our pre-training model achieves new state-of-the-art results in 10 VL understanding tasks and 2 more novel text-to-visual generation tasks. For example, we outperform SOTA models with relative increases of 40.4% R@1 in zero-shot MSR-VTT text-to-video retrieval task and 55.4% in high-resolution dataset LSMDC. The learned VL embedding is also effective in generating visually pleasing and semantically relevant results in text-to-visual editing and super-resolution tasks.

Yanhong Zeng, Huan Yang, Hongyang Chao et al.

We present a new perspective of achieving image synthesis by viewing this task as a visual token generation problem. Different from existing paradigms that directly synthesize a full image from a single input (e.g., a latent code), the new formulation enables a flexible local manipulation for different image regions, which makes it possible to learn content-aware and fine-grained style control for image synthesis. Specifically, it takes as input a sequence of latent tokens to predict the visual tokens for synthesizing an image. Under this perspective, we propose a token-based generator (i.e.,TokenGAN). Particularly, the TokenGAN inputs two semantically different visual tokens, i.e., the learned constant content tokens and the style tokens from the latent space. Given a sequence of style tokens, the TokenGAN is able to control the image synthesis by assigning the styles to the content tokens by attention mechanism with a Transformer. We conduct extensive experiments and show that the proposed TokenGAN has achieved state-of-the-art results on several widely-used image synthesis benchmarks, including FFHQ and LSUN CHURCH with different resolutions. In particular, the generator is able to synthesize high-fidelity images with 1024x1024 size, dispensing with convolutions entirely.

Rui Nie, Huan Yang, Hejuan Peng et al.

Small intestinal capsule endoscopy is the mainstream method for inspecting small intestinal lesions,but a single small intestinal capsule endoscopy will produce 60,000 - 120,000 images, the majority of which are similar and have no diagnostic value. It takes 2 - 3 hours for doctors to identify lesions from these images. This is time-consuming and increase the probability of misdiagnosis and missed diagnosis since doctors are likely to experience visual fatigue while focusing on a large number of similar images for an extended period of time.In order to solve these problems, we proposed a similar wireless capsule endoscope (WCE) image screening method based on structural similarity analysis and the hierarchical clustering of visually salient sub-image blocks. The similarity clustering of images was automatically identified by hierarchical clustering based on the hue,saturation,value (HSV) spatial color characteristics of the images,and the keyframe images were extracted based on the structural similarity of the visually salient sub-image blocks, in order to accurately identify and screen out similar small intestinal capsule endoscopic images. Subsequently, the proposed method was applied to the capsule endoscope imaging workstation. After screening out similar images in the complete data gathered by the Type I OMOM Small Intestinal Capsule Endoscope from 52 cases covering 17 common types of small intestinal lesions, we obtained a lesion recall of 100% and an average similar image reduction ratio of 76%. With similar images screened out, the average play time of the OMOM image workstation was 18 minutes, which greatly reduced the time spent by doctors viewing the images.