×
Clearly, using scientific names in prompts yields poor zero-shot species recognition performance because pretrained VLMs do not necessarily see scientific names. Additionally using descriptions improves marginally. In contrast, simply translating them to common English names significantly boost performance.
Dec 6, 2023
Oct 15, 2023 · We find that common names are more likely to be included in CLIP's training set, and prompting them achieves 2\sim5 times higher accuracy on ...
We find that common names are more likely to be included in CLIP's training set, and prompting them achieves 2 ∼ 5 times higher accuracy on benchmarking ...
People also ask
For instance, Parashar et al. [40] enhanced CLIP's zero-shot performance by substituting scientific CoI names, like "Rosa", with common English names, such as " ...
Prompting Scientific Names for Zero-Shot Species Recognition · Shubham Parashar, Zhiqiu Lin, Yanan Li, Shu Kong. Published: 07 Oct 2023, Last Modified: 01 Dec ...
Zero-shot learning (ZSL) aims to recognise objects whose instances have not yet been seen during training, based on semantic knowledge, e.g., attributes ( ...
Oct 15, 2023 · Trained on web-scale image-text pairs, Vision-Language Models (VLMs) such as CLIP can recognize images of common objects in a zero-shot ...
Oct 17, 2023 · Hello everyone, I am delighted to announce that our EMNLP 2023 submission - "Prompting Scientific Names for Zero-Shot Species Recognition" ...
Prompting scientific names for zero-shot species recognition. S Parashar, Z Lin, Y Li, S Kong. arXiv preprint arXiv:2310.09929, 2023. 7, 2023. Long-tailed 3d ...
We find that common names are more likely to be included in CLIP's training set, and prompting them achieves 2 ∼ 5 times higher accuracy on benchmarking ...