FocusCLIP: Multimodal Subject-Level Guidance for Zero-Shot Transfer in Human-Centric Tasks

sellArticle sellPreprint sellCLIP sellZero-Shot Learning sellHuman-Centric Tasks sellMultimodal Learning sellPose Descriptions

FocusCLIP: Multimodal Subject-Level Guidance for Zero-Shot Transfer in Human-Centric Tasks

We propose FocusCLIP, integrating subject-level guidance--a specialized mechanism for target-specific supervision--into the CLIP framework for improved zero-shot transfer on human-centric tasks. Our novel contributions enhance CLIP on both the vision and text sides. On the vision side, we incorporate ROI heatmaps emulating human visual attention mechanisms to emphasize subject-relevant image regions. On the text side, we introduce human pose descriptions to provide rich contextual information. For human-centric tasks, FocusCLIP is trained with images from the MPII Human Pose dataset. The proposed approach surpassed CLIP by an average of 8.61% across five previously unseen datasets covering three human-centric tasks. FocusCLIP achieved an average accuracy of 33.65% compared to 25.04% by CLIP. We observed a 3.98% improvement in activity recognition, a 14.78% improvement in age classification, and a 7.06% improvement in emotion recognition. Moreover, using our proposed single-shot LLM prompting strategy, we release a high-quality MPII Pose Descriptions dataset to encourage further research in multimodal learning for human-centric tasks. Furthermore, we also demonstrate the effectiveness of our subject-level supervision on non-human-centric tasks. FocusCLIP shows a 2.47% improvement over CLIP in zero-shot bird classification using the CUB dataset. Our findings emphasize the potential of integrating subject-level guidance with general pretraining methods for enhanced downstream performance.

Read Full-Text

  1. We introduce FocusCLIP, a CLIP variant with a third heatmap modality.
  2. FocusCLIP is pretrained on images of people and their body pose descriptions.
  3. FocusCLIP outperforms CLIP on zero-shot classification of activities, emotions, and age in images.
  4. We release our MPII Human Pose Descriptions dataset.

How to Cite

If you find this useful, please include the following citation in your work:

  title={FocusCLIP: Multimodal Subject-Level Guidance for Zero-Shot Transfer in Human-Centric Tasks},
  author={Khan, Muhammad Saif Ullah and Naeem, Muhammad Ferjad and Tombari, Federico and Van Gool, Luc and Stricker, Didier and Afzal, Muhammad Zeshan},
  journal={arXiv preprint arXiv:2403.06904},