Iterative Class Prototype Calibration for Transductive Zero-Shot Learning

Yang, Hai-Rui; Sun, Baoli; Li, Baopu; Yang, Caifei; Wang, Zhihui; Chen, Jenhui; Wang, Lei; Li, Haojie

doi:10.1109/tcsvt.2022.3209209

Cited by 15 publications

(7 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, because there is no knowledge of the unseen data, the inductive ZSL heavily relies on the quality of the auxiliary information, making it challenging to overcome performance bottleneck. Transductive Zero-Shot Learning As a concession of inductive ZSL, TZSL uses test-time unseen data to improve training [14,42,46]. A representative approach is visual structure constraint (VSC) [40].…”

Section: Related Workmentioning

confidence: 99%

Bi-directional Distribution Alignment for Transductive Zero-Shot Learning

Wang¹,

Hao²,

Mu³

et al. 2023

Preprint

View full text Add to dashboard Cite

Zero-shot learning (ZSL) suffers intensely from the domain shift issue, i.e., the mismatch (or misalignment) between the true and learned data distributions for classes without training data (unseen classes). By learning additionally from unlabelled data collected for the unseen classes, transductive ZSL (TZSL) could reduce the shift but only to a certain extent. To improve TZSL, we propose a novel approach Bi-VAEGAN which strengthens the distribution alignment between the visual space and an auxiliary space. As a result, it can reduce largely the domain shift. The proposed key designs include (1) a bi-directional distribution alignment, (2) a simple but effective L 2 -norm based feature normalization approach, and (3) a more sophisticated unseen class prior estimation. Evaluated by four benchmark datasets, Bi-VAEGAN 1 achieves the new state of the art under both the standard and generalized TZSL settings.

show abstract

Section: Related Workmentioning

confidence: 99%

Bi-directional Distribution Alignment for Transductive Zero-Shot Learning

Wang¹,

Hao²,

Mu³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Our FreeU framework exhibits seamless adaptability when integrated with existing diffusion models, encompassing applications like text-to-image generation and textto-video generation. We conduct a comprehensive experimental evaluation of our approach, employing Stable Diffusion [26], DreamBooth [27], ReVersion [12], Mod-elScope [20], and Rerender [34] as our foundational models for benchmark comparisons. By employing FreeU during the inference phase, these models indicate a discernible enhancement in the quality of generated outputs.…”

Section: Denoising Generated Image Low Frequency High Frequencymentioning

confidence: 99%

“…To assess the effectiveness of the proposed FreeU, we systematically conduct a series of experiments, aligning our benchmarks with state-of-the-art methods such as Stable Diffusion [26], DreamBooth [27], ModelScope [20], and Rerender [34]. Importantly, our approach seamlessly integrates with these established methods without imposing any additional computational overhead associated with supple-mentary training or fine-tuning.…”

Section: Implementation Detailsmentioning

confidence: 99%

“…Furthermore, we evaluated FreeU's impact on Rerender [34], a diffusion model tailored for zero-shot textguided video-to-video translations. Fig.…”

Section: Downstream Tasksmentioning

confidence: 99%

“…12 depicts the results: clear improvements in the detail and realism of synthesized videos. For instance, when provided with the prompt "A dog wearing sunglasses" and an input video, Rerender [34] initially produces a dog video with artifacts related to the "sunglasses". However, the incorporation of FreeU successfully eliminates such artifacts, resulting in a refined output.…”

Section: Downstream Tasksmentioning

confidence: 99%

See 2 more Smart Citations