BaFTA: Backprop-Free Test-Time Adaptation For Zero-Shot Vision-Language Models

Large-scale pretrained vision-language models like CLIP have demonstrated remarkable zero-shot image classification capabilities across diverse domains. To enhance CLIP's performance while preserving the zero-shot paradigm, various test-time prompt tuning methods have been introduced to refine class embeddings through unsupervised learning objectives during inference. However, these methods often encounter challenges in selecting appropriate learning rates to prevent collapsed training in the absence of validation data during test-time adaptation. In this study, we propose a novel backpropagation-free algorithm BaFTA for test-time adaptation of vision-language models. Instead of fine-tuning text prompts to refine class embeddings, our approach directly estimates class centroids using online clustering within a projected embedding space that aligns text and visual embeddings. We dynamically aggregate predictions from both estimated and original class embeddings, as well as from distinct augmented views, by assessing the reliability of each prediction using R\'enyi Entropy. Through extensive experiments, we demonstrate that BaFTA consistently outperforms state-of-the-art test-time adaptation methods in both effectiveness and efficiency.

GaitSTR: Gait Recognition with Sequential Two-stream Refinement

Gait recognition aims to identify a person based on their walking sequences, serving as a useful biometric modality as it can be observed from long distances without requiring cooperation from the subject. In representing a person's walking sequence, silhouettes and skeletons are the two primary modalities used. Silhouette sequences lack detailed part information when overlapping occurs between different body segments and are affected by carried objects and clothing. Skeletons, comprising joints and bones connecting the joints, provide more accurate part information for different segments; however, they are sensitive to occlusions and low-quality images, causing inconsistencies in frame-wise results within a sequence. In this paper, we explore the use of a two-stream representation of skeletons for gait recognition, alongside silhouettes. By fusing the combined data of silhouettes and skeletons, we refine the two-stream skeletons, joints, and bones through self-correction in graph convolution, along with cross-modal correction with temporal consistency from silhouettes. We demonstrate that with refined skeletons, the performance of the gait recognition model can achieve further improvement on public gait recognition datasets compared with state-of-the-art methods without extra annotations.

Large Language Models are Good Prompt Learners for Low-Shot Image Classification

Low-shot image classification, where training images are limited or inaccessible, has benefited from recent progress on pre-trained vision-language (VL) models with strong generalizability, e.g. CLIP. Prompt learning methods built with VL models generate text features from the class names that only have confined class-specific information. Large Language Models (LLMs), with their vast encyclopedic knowledge, emerge as the complement. Thus, in this paper, we discuss the integration of LLMs to enhance pre-trained VL models, specifically on low-shot classification. However, the domain gap between language and vision blocks the direct application of LLMs. Thus, we propose LLaMP, Large Language Models as Prompt learners, that produces adaptive prompts for the CLIP text encoder, establishing it as the connecting bridge. Experiments show that, compared with other state-of-the-art prompt learning methods, LLaMP yields better performance on both zero-shot generalization and few-shot image classification, over a spectrum of 11 datasets.

CaesarNeRF: Calibrated Semantic Representation for Few-shot Generalizable Neural Rendering

Generalizability and few-shot learning are key challenges in Neural Radiance Fields (NeRF), often due to the lack of a holistic understanding in pixel-level rendering. We introduce CaesarNeRF, an end-to-end approach that leverages scene-level CAlibratEd SemAntic Representation along with pixel-level representations to advance few-shot, generalizable neural rendering, facilitating a holistic understanding without compromising high-quality details. CaesarNeRF explicitly models pose differences of reference views to combine scene-level semantic representations, providing a calibrated holistic understanding. This calibration process aligns various viewpoints with precise location and is further enhanced by sequential refinement to capture varying details. Extensive experiments on public datasets, including LLFF, Shiny, mip-NeRF 360, and MVImgNet, show that CaesarNeRF delivers state-of-the-art performance across varying numbers of reference views, proving effective even with a single reference image. The project page of this work can be found at https://haidongz-usc.github.io/project/caesarnerf.

ShARc: Shape and Appearance Recognition for Person Identification In-the-wild

Identifying individuals in unconstrained video settings is a valuable yet challenging task in biometric analysis due to variations in appearances, environments, degradations, and occlusions. In this paper, we present ShARc, a multimodal approach for video-based person identification in uncontrolled environments that emphasizes 3-D body shape, pose, and appearance. We introduce two encoders: a Pose and Shape Encoder (PSE) and an Aggregated Appearance Encoder (AAE). PSE encodes the body shape via binarized silhouettes, skeleton motions, and 3-D body shape, while AAE provides two levels of temporal appearance feature aggregation: attention-based feature aggregation and averaging aggregation. For attention-based feature aggregation, we employ spatial and temporal attention to focus on key areas for person distinction. For averaging aggregation, we introduce a novel flattening layer after averaging to extract more distinguishable information and reduce overfitting of attention. We utilize centroid feature averaging for gallery registration. We demonstrate significant improvements over existing state-of-the-art methods on public datasets, including CCVID, MEVID, and BRIAR.

ReCLIP: Refine Contrastive Language Image Pre-Training with Source Free Domain Adaptation

Large-scale Pre-Training Vision-Language Model such as CLIP has demonstrated outstanding performance in zero-shot classification, e.g. achieving 76.3% top-1 accuracy on ImageNet without seeing any example, which leads to potential benefits to many tasks that have no labeled data. However, while applying CLIP to a downstream target domain, the presence of visual and text domain gaps and cross-modality misalignment can greatly impact the model performance. To address such challenges, we propose ReCLIP, the first source-free domain adaptation method for vision-language models, which does not require any source data or target labeled data. ReCLIP first learns a projection space to mitigate the misaligned visual-text embeddings and learns pseudo labels, and then deploys cross-modality self-training with the pseudo labels, to update visual and text encoders, refine labels and reduce domain gaps and misalignments iteratively. With extensive experiments, we demonstrate ReCLIP reduces the average error rate of CLIP from 30.17% to 25.06% on 22 image classification benchmarks.

CAILA: Concept-Aware Intra-Layer Adapters for Compositional Zero-Shot Learning

Compositionality, the ability to combine existing concepts and generalize towards novel compositions, is a key functionality for intelligent entities. Here, we study the problem of Compositional Zero-Shot Learning (CZSL), which aims at recognizing novel attribute-object compositions. Recent approaches build their systems on top of large-scale Vision-Language Pre-trained (VLP) models, e.g. CLIP, and observe significant improvements. However, these methods treat CLIP as a black box and focus on pre- and post-CLIP operations. Here, we propose to dive deep into the architecture and insert adapters, a parameter-efficient technique proven to be effective among large language models, to each CLIP encoder layer. We further equip adapters with concept awareness so that concept-specific features of "object", "attribute" and "composition" can be extracted. We name our method CAILA, Concept-Aware Intra-Layer Adapters. Quantitative evaluations performed on three popular CZSL datasets, MIT-States, C-GQA, and UT-Zappos, reveal that CAILA achieves double-digit relative improvements against the current state-of-the-art on all benchmarks.

GaitRef: Gait Recognition with Refined Sequential Skeletons

Identifying humans with their walking sequences, known as gait recognition, is a useful biometric understanding task as it can be observed from a long distance and does not require cooperation from the subject. Two common modalities used for representing the walking sequence of a person are silhouettes and joint skeletons. Silhouette sequences, which record the boundary of the walking person in each frame, may suffer from the variant appearances from carried-on objects and clothes of the person. Framewise joint detections are noisy and introduce some jitters that are not consistent with sequential detections. In this paper, we combine the silhouettes and skeletons and refine the framewise joint predictions for gait recognition. With temporal information from the silhouette sequences. We show that the refined skeletons can improve gait recognition performance without extra annotations. We compare our methods on four public datasets, CASIA-B, OUMVLP, Gait3D and GREW, and show state-of-the-art performance.

CAT-NeRF: Constancy-Aware Tx$^2$Former for Dynamic Body Modeling

This paper addresses the problem of human rendering in the video with temporal appearance constancy. Reconstructing dynamic body shapes with volumetric neural rendering methods, such as NeRF, requires finding the correspondence of the points in the canonical and observation space, which demands understanding human body shape and motion. Some methods use rigid transformation, such as SE(3), which cannot precisely model each frame's unique motion and muscle movements. Others generate the transformation for each frame with a trainable network, such as neural blend weight field or translation vector field, which does not consider the appearance constancy of general body shape. In this paper, we propose CAT-NeRF for self-awareness of appearance constancy with Tx$^2$Former, a novel way to combine two Transformer layers, to separate appearance constancy and uniqueness. Appearance constancy models the general shape across the video, and uniqueness models the unique patterns for each frame. We further introduce a novel Covariance Loss to limit the correlation between each pair of appearance uniquenesses to ensure the frame-unique pattern is maximally captured in appearance uniqueness. We assess our method on H36M and ZJU-MoCap and show state-of-the-art performance.

Efficient Feature Distillation for Zero-shot Detection

The large-scale vision-language models (e.g., CLIP) are leveraged by different methods to detect unseen objects. However, most of these works require additional captions or images for training, which is not feasible in the context of zero-shot detection. In contrast, the distillation-based method is an extra-data-free method, but it has its limitations. Specifically, existing work creates distillation regions that are biased to the base categories, which limits the distillation of novel category information and harms the distillation efficiency. Furthermore, directly using the raw feature from CLIP for distillation neglects the domain gap between the training data of CLIP and the detection datasets, which makes it difficult to learn the mapping from the image region to the vision-language feature space - an essential component for detecting unseen objects. As a result, existing distillation-based methods require an excessively long training schedule. To solve these problems, we propose Efficient feature distillation for Zero-Shot Detection (EZSD). Firstly, EZSD adapts the CLIP's feature space to the target detection domain by re-normalizing CLIP to bridge the domain gap; Secondly, EZSD uses CLIP to generate distillation proposals with potential novel instances, to avoid the distillation being overly biased to the base categories. Finally, EZSD takes advantage of semantic meaning for regression to further improve the model performance. As a result, EZSD achieves state-of-the-art performance in the COCO zero-shot benchmark with a much shorter training schedule and outperforms previous work by 4% in LVIS overall setting with 1/10 training time.