11 November 2024 in IEEE Transactions on Circuits and Systems for Video Technology
In few-shot fine-grained recognition (FS-FGR) tasks, the main challenge is to distinguish novel categories with high intra-class variations and low inter-class differences given scarce training data. Existing studies explore discriminative features through a compact network to avoid overfitting, while they achieve marginal performance gain owing to the limited representation capability. Motivated by the significant progress of the vision foundation model, we introduce it to describe visual attributes and boost the performance of the compact feature extractor. A few-shot fine-grained recognition method with Variational Feature Imitation Conditioned on Visual Descriptions, VFI-CVD for short, has been proposed in this paper. It simultaneously exploits the pre-trained knowledge from a vision foundation model and the expert knowledge mined by a feature extractor. Specifically, the intra-class variations shared across object categories are encoded into a common distribution thus we can augment features by sampling latent variables. To enhance the learning of intra-class variations, a condition exchange strategy (CES) is put forward to interact the knowledge between samples through feature cross-imitation. In the inference stage, the learned knowledge is further integrated through the joint prediction of visual descriptions and cross-imitated features. Comprehensive experimental results on four fine-grained benchmark datasets show that the proposed VFI-CVD achieves state-of-the-art performance, e.g., 90.37% under the 5-way 1-shot setting on CUB-200-2011. It surpasses existing methods by a large margin, especially in the challenging 30-way recognition tasks and cross-domain evaluation. The source code is publicly available: https://github.com/Lx-zjwf/VFI-CVD.
10 June 2026 in IEEE Transactions on Circuits and Systems for Video Technology
3D human pose estimation (HPE) from monocular 2D videos faces a critical trade-off between joint position accuracy and structural plausibility. Existing methods relying on joint coordinates suffer from implausible bone lengths, while bone vector-based approaches incur error accumulation, degrading position accuracy. To address this challenge, we propose a Decouple-then-Collaborate framework that exploits the complementary strengths of joint coordinates and bone vectors. Our approach decouples the learning of the two representations—joint coordinates for positional accuracy and bone vectors for structural consistency—and then aligns their feature spaces to fuse these advantages. Specifically, we first train two separate branches, each equipped with a motion enhancement module to learn motion-aware features. A feature transfer network then maps high-precision joint features into the structurally stable bone feature space, allowing both representations to be obtained with a single backbone forward pass during inference. Finally, a geometric calibration module integrates both predictions under explicit bone-length constraints. Extensive experiments on Human3.6M and MPI-INF-3DHP demonstrate that our method not only achieves state-of-the-art accuracy but also significantly improves structural stability and robustness to input noise, providing a practical solution for accurate and stable 3D pose estimation in video.
1 January 2026 in IEEE Transactions on Robotics