High-Resolution Volumetric Reconstruction for Clothed Humans

We present a novel method for reconstructing clothed humans from a sparse set of, e.g., 1 to 6 RGB images. Despite impressive results from recent works employing deep implicit representation, we revisit the volumetric approach and demonstrate that better performance can be achieved with proper system design. The volumetric representation offers significant advantages in leveraging 3D spatial context through 3D convolutions, and the notorious quantization error is largely negligible with a reasonably large yet affordable volume resolution, e.g., 512. To handle memory and computation costs, we propose a sophisticated coarse-to-fine strategy with voxel culling and subspace sparse convolution. Our method starts with a discretized visual hull to compute a coarse shape and then focuses on a narrow band nearby the coarse shape for refinement. Once the shape is reconstructed, we adopt an image-based rendering approach, which computes the colors of surface points by blending input images with learned weights. Extensive experimental results show that our method significantly reduces the mean point-to-surface (P2S) precision of state-of-the-art methods by more than 50% to achieve approximately 2mm accuracy with a 512 volume resolution. Additionally, images rendered from our textured model achieve a higher peak signal-to-noise ratio (PSNR) compared to state-of-the-art methods.

Cluster Contrast for Unsupervised Person Re-Identification

Unsupervised person re-identification (re-ID) attracts increasing attention due to its practical applications in industry. State-of-the-art unsupervised re-ID methods train the neural networks using a memory-based non-parametric softmax loss. They store the pre-computed instance feature vectors inside the memory, assign pseudo labels to them us-ing clustering algorithm, and compare the query instances to the cluster using a form of contrastive loss. During training, the instance feature vectors are updated. How-ever, due to the varying cluster size, the updating progress for each cluster is inconsistent. To solve this problem, we present Cluster Contrast which stores feature vectors and computes contrast loss in the cluster level. We demonstrate that the inconsistency problem for cluster feature representation can be solved by the cluster-level memory dictionary.By straightforwardly applying Cluster Contrast to a standard unsupervised re-ID pipeline, it achieves considerable improvements of 9.5%, 7.5%, 6.6% compared to state-of-the-art purely unsupervised re-ID methods and 5.1%, 4.0%,6.5% mAP compared to the state-of-the-art unsupervised domain adaptation re-ID methods on the Market, Duke, andMSMT17 datasets.Our source code is available at https://github.com/alibaba/cluster-contrast.

Spherical Text Embedding

Unsupervised text embedding has shown great power in a wide range of NLP tasks. While text embeddings are typically learned in the Euclidean space, directional similarity is often more effective in tasks such as word similarity and document clustering, which creates a gap between the training stage and usage stage of text embedding. To close this gap, we propose a spherical generative model based on which unsupervised word and paragraph embeddings are jointly learned. To learn text embeddings in the spherical space, we develop an efficient optimization algorithm with convergence guarantee based on Riemannian optimization. Our model enjoys high efficiency and achieves state-of-the-art performances on various text embedding tasks including word similarity and document clustering.

CatE: Category-Name GuidedWord Embedding

Unsupervised word embedding has benefited a wide spectrum of NLP tasks due to its effectiveness of encoding word semantics in distributed word representations. However, unsupervised word embedding is a generic representation, not optimized for specific tasks. In this work, we propose a weakly-supervised word embedding framework, CatE. It uses category names to guide word embedding and effectively selects category representative words to regularize the embedding space where the categories are well separated. Experiments show that our model outperforms unsupervised word embedding models significantly on both document classification and category representative words retrieval tasks.