CLIP4Sketch Synthetic Dataset
收藏资源简介:
CLIP4Sketch合成数据集由国际信息技术研究所-海得拉巴创建,旨在通过扩散模型增强素描与头像匹配的性能。该数据集包含245,376张素描图像,对应27,264个独特身份,具有四种手绘和四种软件生成风格。数据集的创建过程利用了去噪扩散概率模型(DDPMs),结合CLIP和Adaface嵌入以及文本描述作为条件,生成具有身份和风格控制的素描图像。该数据集主要应用于法医素描与头像匹配领域,旨在解决现有数据集稀缺和模态差异问题,提升面部识别系统的准确性。
CLIP4Sketch synthetic dataset was developed by the International Institute of Information Technology, Hyderabad, with the goal of enhancing the performance of sketch-to-face matching via diffusion models. This dataset contains 245,376 sketch images corresponding to 27,264 unique identities, featuring four hand-drawn and four software-generated sketch styles. The dataset was constructed using denoising diffusion probabilistic models (DDPMs), with CLIP and Adaface embeddings as well as textual descriptions used as conditioning signals to generate sketch images with controllable identity and style attributes. This dataset is primarily utilized in the field of forensic sketch-to-face matching, aiming to address the issues of scarcity of existing datasets and cross-modal discrepancy, thereby improving the accuracy of facial recognition systems.

- 1CLIP4Sketch: Enhancing Sketch to Mugshot Matching through Dataset Augmentation using Diffusion Models国际信息技术研究所-海得拉巴 · 2024年



